Interpretation of gene and genomic variants by integrated calculation and experiment of deep mutation learning framework

By combining a deep mutation learning system with high-throughput molecular measurements and single-cell technology, the accuracy and scalability issues of assessing the phenotypic impact of genotype variants have been resolved. This enables multidimensional classification of molecular variants and accurate prediction of their phenotypic impact, supporting genomic diagnostics and life science applications.

CN119028454APending Publication Date: 2024-11-26LABORATORY CORPORATION OF AMERICA HOLDINGS INC
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202410955984.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2018-03-08
Filing Date
2018-06-19
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively assess the impact of genotype variants on cellular and organismal phenotypes, particularly in underdeveloped functional elements and disease-associated genes, resulting in insufficient accuracy and extensibility in variant classification and interpretation.

Method used

We employ a deep mutation learning (DML) system that combines high-throughput molecular measurements, single-cell manipulation, molecular biology, and computational modeling techniques. We use statistical learning methods to perform multidimensional classification of molecular variants in the genome and utilize biological signals from single cells and cell populations to identify and predict phenotypic effects.

Benefits of technology

It enables robust and scalable assessment of molecular variants, improves the accuracy of predicting phenotypic effects in different variant types, biophysical processes, and molecular functional contexts, and supports genomic diagnostics in precision medicine and life sciences.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119028454A_ABST
    Figure CN119028454A_ABST
Patent Text Reader

Abstract

Disclosed herein are system, method, and computer program product embodiments for determining the phenotypic impact of identified molecular variants within a biological sample. Embodiments include receiving molecular variants associated with functional elements within a model system. Embodiments then determine molecular scores associated with the model system. Embodiments then determine a molecular signal and a population signal associated with the molecular variant based on the molecular score. Embodiments then determine functional scores for the molecular variants based on statistical learning. Embodiments then derive an evidence score for the molecular variant based on the functional score. Embodiments then determine a phenotypic impact of the molecular variant based on the functional score or evidence score.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] This application is a divisional application of the invention patent application of the same name 201880050685.7 filed on June 19, 2018.

[0002]

Overview

[0003] Understanding the impact of genotypic (e.g., sequence) variation within functional elements of the genome, such as protein-coding genes, noncoding genes, and regulatory elements, is critical for a variety of life science applications. Currently, almost half of all disease-associated genes in the general population contain a higher number of uncharacterized variants than variants of known clinical significance. This poses a significant challenge for both diagnostic and screening tests evaluating gene sequences and genomic sequences ( Landrum et al., 2015; Lek et al., 2016 Large numbers of novel variants of unknown clinical significance are characteristic of nearly all genes (e.g., for both germline and somatic variants in a population), affecting even the most frequently tested genes. For example, a test evaluating a panel of genes for cancer susceptibility mutations reported finding up to 95 uncharacterized variants for each known pathogenic variant ( Maxwell et al., 2016 Year ). Thus, predicting the phenotypic (e.g., cellular, organismal, clinical, or other) consequences of genotypic variants is an obstacle to utilizing genetic and genomic information in a wide range of clinical settings.

[0004] Genotypic (e.g., sequence) variants within genomic encoded functional elements can affect a variety of biophysical processes, thereby altering the unique molecular functions within each element and leading to a variety of clinical and nonclinical phenotypes. For example, in the established tumor suppressor protein encoding gene phosphatase-tensin homolog (PTEN), genotypic variants affecting transcription (e.g., –903G>A, –975G>C, and –1026C>A), genotypic variants affecting protein stability (e.g., C136R), genotypic variants affecting phosphatase catalytic activity (e.g., C124S, H93R), and genotypic variants affecting substrate recognition (e.g., G129E) are all associated with Cowden syndrome (CS), presenting a high risk of breast cancer, thyroid cancer, endometrial cancer, renal cancer, colorectal cancer, and melanoma ( Heikkinen et al., 2011; He et al., 2013; Myers et al., 1997; Myers et al., 1998 Year ). Variants affecting the same biophysical processes and molecular functions can lead to comorbidity between different disorders, such as PTEN variants affecting phosphatase activity (e.g., H93R), which are additionally associated with autism spectrum disorder (ASD) ( Johnston and Raines, 2015 ), leading to frequent comorbidity between ASD and cancer ( Markkanen et al., 2016). In addition, variants within functional elements that affect different biophysical processes and molecular mechanisms can present with patterned, differentiated clinical and nonclinical phenotypes. Mutations in the lamin A / C gene (LMNA) cause a total of more than 15 diseases, which are collectively referred to as "laminopathies", including A-EDMD (autosomal Emerging-Derrickson muscular dystrophy), DCM (dilated cardiomyopathy), LGMD1B (limb-girdle muscular dystrophy 1B), L-CMD (LMNA-related congenital muscular dystrophy), FPLD2 (familial partial lipodystrophy 2), HGPS (Hutchinson-Gilford progeria syndrome), atypical WRN (Werner syndrome), MAD (mandibular dysplasia), and CMT2B (Charcot-Marie-Tooth disorder type 2B) ( Scharner et al., 2010 In LMNA, genotypic (e.g., sequence) variants leading to HGPS generate a cryptic splice site donor in lamin A-specific exon 11, which results in a truncated form of lamin A, whereas variants leading to FPLD2 alter the surface charge of the Ig-like domain and do not change the crystal structure of the mutant protein ( Scharner et al., 2010 ). Therefore, the complexity of untangling the genotype-phenotype relationships among the various variant types, functional elements, and molecular systems and cellular effects is an outstanding challenge for the robust and scalable interpretation of phenotypic outcomes of variants discovered in clinical and nonclinical genetic and genomic tests.

[0005] Indeed, assessing the significance of genotypic (e.g., sequence) variants can be a complex and challenging task. As recently as 2015, a survey of variant classifications revealed that up to 17% (e.g., 2,229 / 12,895) of variants were inconsistently classified among classification submitters (Rehm et al., 2015). Among clinical testing laboratories, although specific recommendations have been shown to improve interlaboratory agreement to 71%, agreement in interpretation has been measured as low as 34% (Amendola et al., 2016).

[0006] With over 5,300 genes evaluated by genetic tests in the market (e.g., according to the NCBI Genetic Test Registry), scalable solutions for interpreting (e.g., classifying) genotype (e.g., sequence) variants in a variety of genes, diseases, and settings (e.g., clinical and non-clinical) are critical to the work of the precision medicine and life sciences industries. There are over 14,000,000 possible (e.g., unique) molecular variants within the subset of molecular variants corresponding to single nucleotide variants (SNVs), within the subset of coding sequences, and within the subset of protein-coding genes in the clinical testing market, so effective solutions for molecular variant classification need to be robust and scalable.

[0007] Although multiple strategies are currently available to identify the phenotypic impact of molecular variants (including but not limited to family isolation, functional assays, and case-control studies), only computational predictors of variant effects provide supporting evidence at the required scale. Indeed, an analysis of clinical variant classifications from practitioners according to the joint American College of Medical Genetics and Genomics (ACMG) and Association for Molecular Pathology (AMP) guidelines for clinical variant interpretation showed that approximately 50% of clinical variant classifications rely on the use of computational predictors of variant effects. However, despite their widespread use, benchmarking studies have shown that computational prediction algorithms for variant effects (such as SIFT, PolyPhen (v2), GERP++, Condel, CADD, REVEL, etc.) have significantly low performance, with area under the curve (AUC) ranging from 0.52-0.75 (Mahmood et al., 2017).

[0008] Direct determination of molecular function can provide a basis for accurate interpretation of the clinical and nonclinical impact of genotypic (e.g., sequence) variants ( Shendure and Fields, 2016; Araya and Fowler, 2011 To date, a variety of assays have been designed to directly assess the effects of variants on various molecular functions. However, existing methods require prior knowledge or assumptions about the mechanism of action of the variant associated with the clinical (and nonclinical) phenotype being studied to define the molecular function to be assayed ( Shendure and Fields, 2016) . These methods are generally limited to capturing and deriving information about the effects of variants that affect the specific molecular function being assayed, thereby placing restrictions on the types of variants that can be assayed on a large scale, the types of molecular functions, and the types of functional elements and genes. Thus, while, for example, a phosphatase assay may nominate (e.g., rule in) potential disease associations for variants that affect the catalytic activity of a PTEN tumor suppressor, such an assay may not rule out (e.g., rule out) potential disease associations for variants that affect protein stability, since these variants may increase the risk of developing a disease without an apparent defect in catalytic activity. Conversely, while, for example, a protein stability assay may nominate (e.g., rule in) potential disease associations for variants that cause a stability defect in a PTEN tumor suppressor, such an assay may not rule out (e.g., rule out) potential disease associations for variants that affect catalytic activity. The potential need for prior knowledge or hypotheses about mechanisms of action (and therefore the associated molecular functions to be assayed) may limit the application of these methods to well-characterized functional elements (e.g., genes) and phenotypes, which may prevent their application to poorly understood disease-associated genes.

[0009] Based on the technology of high-throughput DNA sequencing platforms, recently developed large-scale functional assays (such as Deep Mutational Scanning (DMS), HITS-KIN, RNA-MAP, etc.) have enabled comprehensive or near-comprehensive coverage of possible sequence variants in different sequence categories, including single nucleotide variants (SNVs) and non-synonymous variants (NSVs, missense variants) in coding elements, non-coding elements, and regulatory elements ( Fowler et al., 2010; Araya et al., 2012; Guenther et al., 2013; Buenrostro et al., 2014; Kelsic et al., 2016; Patwardhan et al., 2009). Such methods can serve as the basis for robust, statistically validated interpretations of the effects of molecular variants, such as genotypic (e.g., sequence) variants, on patient phenotypes, including clinical phenotypes, such as increased risk of lipodystrophy and type 2 diabetes (T2D) in patients with PPARG variants, or increased risk of breast and ovarian cancer in patients with BRCA1 variants. Starita et al., 2015; Majithia et al., 2016 ). Although such methods can provide robust variant interpretation in clinical and non-clinical testing settings, these methods may require significant development and customization to determine each molecular function and each functional element. This may limit their practicality as a general, scalable solution to systematically evaluate the clinical and non-clinical consequences of molecular variants (such as genotype (e.g., sequence) variants) in the context of different types of variants, biophysical processes, molecular functions, functional elements, genes, and ultimately pathways. Therefore, a versatile platform and method for variant impact assessment is needed.

Brief Description of the Drawings

[0010] The accompanying drawings are incorporated herein and form a part of the specification.

[0011] Figures 1A-1C Shown are a comprehensive functional assay and computational deep mutational learning (DML) process and system for determining the phenotypic impact of molecular variants, according to some embodiments, and example (e.g., intermediate) data generated by applying the process and system to two genes in the RAS / MAPK family of disease.

[0012] Figure 2A-2B Shown is the performance of a deep mutation learning (DML) process and system in identifying (e.g., binary classification) pathogenic (e.g., pathogenic) and neutral (e.g., benign) molecular variants for germline (e.g., genetic) and somatic disorders in three genes of the RAS / MAPK pathway (HRAS, PTPN11, and MAP2K2), according to some embodiments.

[0013] Figure 3A-3BShown is the performance of a deep mutation learning (DML) pipeline and system in identifying (eg, binary classification) cells that have germline-damaging (eg, pathogenic) or neutral (eg, benign) molecular variants in MAP2K2, according to some embodiments.

[0014] Figure 4 The architecture of a neural network-based denoising autoencoder is shown, which is trained and applied to generate a robust simplified representation of numerator scores according to some embodiments.

[0015] Figure 5 Shown is normalized ERK pathway activation as the ratio of phosphorylated total ERK protein measured by ELISA of cell extracts from H293 cells carrying control, wild-type, and mutant forms of MAP2K2 and PTPN11, according to some embodiments.

[0016] Figure 6 According to some embodiments, a method for performing the multi-dimensional model analysis of the dimensionality reduction model (m) at different cell numbers, read segment depths, and dimensionality reduction models (m) is shown. DR ) and functional model (m F ) case, where optimization is first performed on a (reduced) true set of molecular variants, and a target set including the molecular variants is deployed.

[0017] Figure 7 An example of a method for calculating a phenotypic score according to some embodiments is shown.

[0018] Figure 8 An example of a method for calculating a molecular score according to some embodiments is shown.

[0019] Fig. 9 Methods for calculating molecular signals associated with individual molecular variants are shown according to some embodiments.

[0020] Fig.10 Methods for calculating independent or disjoint estimates of molecular state specificity of molecular signals are shown according to some embodiments.

[0021] Fig.11 Methods for characterizing the distribution of cells with a particular molecular variant across molecular states or phenotypic scores and deriving a population signal are shown according to some embodiments.

[0022] Fig.12An example of a method for identifying high-order molecular signals from low-order molecular signals associated with individual molecular variants using unsupervised learning techniques is shown, according to some embodiments.

[0023] Fig.13 Examples of methods for deriving functional scores and functional classifications by machine learning that associates molecular, phenotypic, or population signals with phenotypic effects of molecular variants by regression and classification techniques are shown according to some embodiments.

[0024] Figures 14A-14B Shown are examples of the performance of methods and systems for binomial classification of molecular variants with two unique phenotypic effects when trained using different numbers of cells, according to some embodiments.

[0025] Fig.15 An example of a method according to some embodiments is shown that allows the use of functional scores and functional classifications from a subset of all possible non-synonymous variants in a protein-coding gene to infer a sequence-function map that describes the functional scores or functional classifications of the possible non-synonymous variants.

[0026] Fig.16 An example of a system and method for reducing the cost and increasing the scope of a DML process for determining the phenotypic impact of a molecular variant through a series of modeling layers is shown in accordance with some embodiments.

[0027] Fig.17 An example of a method according to some embodiments for using machine learning techniques to generate a low-order variant interpretation engine (VIE), which can be gene- and condition-specific, is shown.

[0028] Fig.18 An example of a method for identifying significantly mutated regions (SMRs) and significantly mutated networks (SMNs) according to some embodiments is shown.

[0029] Fig.19 An exemplary computer system is shown that can be used to implement various embodiments.

[0030] In the drawings, like reference numbers generally refer to like or similar elements. Additionally, generally, the leftmost digit(s) of a reference number identifies the drawing in which the reference number first appears. [Specific implementation method]

[0031] Provided herein are systems, devices, apparatus, methods and / or computer program product embodiments, and / or combinations and sub-combinations thereof, for enabling multifunctional, multi-element and multi-gene (e.g., pathway-level) assessment of the phenotypic impact of variants in the context of a wide variety of variant types, biophysical processes, molecular functions, and phenotypic contexts.

[0032] The present disclosure provides system, apparatus, device, method and / or computer program product embodiments that can utilize high-throughput molecular measurement (e.g., next generation sequencing) technology, single-cell manipulation technology, molecular biology technology, computational modeling technology and statistical learning technology, and can perform multifunctional, multi-element and multi-gene (pathway level) assessment of the phenotypic impact of variants in a wide variety of variant types, biophysical processes, molecular functions and phenotypic backgrounds.

[0033] The present disclosure provides system, apparatus, device, method and / or computer program product embodiments for systematically determining and statistically validating one or more phenotypic (e.g., clinical or non-clinical) effects (e.g., pathogenicity, functionality or relative effects) of molecular variants (such as genotypic (e.g., sequence) variants) identified in one or more (e.g., coding or non-coding) functional elements (e.g., protein-coding genes, non-coding genes, molecular domains such as protein or RNA domains, promoters, enhancers, silencers, regulatory binding sites, replication origins, etc.) in a (e.g., nuclear, mitochondrial, etc.) genome or its derived molecules within a biological sample of a subject or a record thereof.

[0034] The present disclosure provides systems, devices, apparatus, methods and / or computer program product embodiments for classifying (or regressing) possible phenotypic effects in a subject based on one or more molecular signals, phenotypic signals or population signals measured in an in vivo or in vitro functional model system. The resulting regression or classification may be referred to as a functional score or functional classification.

[0035] The embodiments herein depart from existing computational or functional evidence-based systems for molecular variant classification, as used, for example, in clinical genetic diagnosis and genomic diagnostics.

[0036] First, although existing computational methods and systems for variant classification rely on a variety of population, evolutionary, physicochemical, structural and / or molecular annotations and properties for variant classification, existing computational methods and systems do not employ information about the effects of molecular variants on cell biology. As a result, such computational methods cannot capture phenotypic effects that act through changes in intracellular molecular properties or changes in cell populations and cellular heterogeneity.

[0037] Second, existing large-scale functional assays and solutions capable of measuring the activity of thousands of molecular variants provide activity measurements along a single dimension for each molecular variant and typically require a priori knowledge or hypotheses about the mechanism of action by which the molecular variant exerts its phenotypic effect.

[0038] Due to these limitations, although conventional computational methods and systems for variant classification can access data covering a variety of annotations and parameters, the performance of these conventional methods in the classification (and regression) tasks of molecular variant phenotypic effects is significantly poor. Similarly, these conventional methods require prior knowledge or assumptions about the mechanism of action (and therefore the associated molecular functions to be determined), which limits their application to fully characterized functional elements (such as genes). This further excludes its application on disease-associated genes that are rarely understood. Finally, these conventional methods require a lot of development and customization to determine each molecular function and each functional element.

[0039] In the embodiments herein, the technical solution to overcome these technical problems relates to such a data structure, which provides a multidimensional representation of cells and cell populations with a specific genotype (e.g., a molecular variant) in one or more functional elements (e.g., a gene) and in one or more backgrounds (e.g., a cell type, a drug treatment, a genotype background). Such a data structure enables systems and methods for statistical learning to achieve improved accuracy in classification tasks related to the phenotypic effects of genotypes (e.g., molecular variants or combinations thereof).

[0040] By acquiring hundreds to tens of thousands (~10 2 –10 4 ) molecular measurements, tens to thousands (~10 1 –10 3 ) molecular profiles of model systems, constructing thousands (~10) of 3 ) molecular profiles of multiple molecular variants, the embodiments herein enable robust, scalable, multidimensional classification of molecular variants (and combinations thereof) across a variety of functional elements and phenotypes.

[0041] like Figure 1AAs shown in, the embodiment of the present disclosure integrates the variant library generation 102 method for high-throughput mutagenesis and the cell library generation 104 method for cell engineering technology to generate a model system (e.g., cell) with unique molecular variants in the target functional element (e.g., gene). The embodiment provides processing, single cell capture, library preparation, sequencing 106 methods, which utilize cell technology, molecular biology technology, and genomics technology as well as technology for processing and capturing model systems, preparation of molecular entity libraries, and technology for measuring multiple molecular entities (e.g., transcripts) within the model system. The embodiment provides mapping, normalization 108 bioinformatics technology, computational biology technology, and statistical technology for mapping, quantifying, and normalizing the association between molecular variants, model systems, and molecular entities within each model system. The embodiment provides feature selection, dimensionality reduction 110 and background annotation, training, classification 112 statistical (e.g., machine) learning technology, distributed and high-performance computing technology, systems biology technology, population and clinical genomics technology for label generation, feature selection, dimensionality reduction, training, and classification of molecular variants.

[0042] In some embodiments, the present disclosure describes the use of Figure 1A These methods and technology series determine the phenotypic impact of the molecular variants identified in the biological sample. In some embodiments, the present disclosure describes the introduction of molecular variants into one or more functional elements in the model system. The model system may include a single cell, a cell compartment, a subcellular compartment, or a synthetic compartment. In some embodiments, the present disclosure describes the determination of the molecular score or phenotypic score of a single cell, a cell compartment, a subcellular compartment, or a synthetic compartment. In some embodiments, the present disclosure describes the identification of molecular variants in a single cell, a cell compartment, a subcellular compartment, or a synthetic compartment. As understood by those of ordinary skill in the art, various methods can be used to identify molecular variants in a single cell, a cell compartment, a subcellular compartment, or a synthetic compartment. This can be based on molecular measurements of a single cell, a cell compartment, a subcellular compartment, or a synthetic compartment. In some embodiments, the present disclosure describes the determination of molecular signals or phenotypic signals associated with each molecular variant based on the molecular score or phenotypic score from a single cell, a cell compartment, a subcellular compartment, or a synthetic compartment associated with a specific molecular variant. In some embodiments, the present disclosure describes determining a population signal associated with a molecular variant based on a molecular score or a phenotypic score for a single cell, cellular compartment, subcellular compartment, or synthetic compartment associated with a particular molecular variant.

[0043] In some embodiments, the present disclosure describes determining a functional score or functional classification of a molecular variant by applying a statistical (e.g., machine) learning method, wherein the learning method associates a molecular signal, a phenotypic signal, or a population signal with the phenotypic impact of a molecular variant. In some embodiments, the present disclosure describes determining an evidence score or an evidence classification of a molecular variant based on a functional score, a functional classification, a prediction score, a prediction classification, a hotspot score, or a hotspot classification. In some embodiments, the present disclosure describes determining the phenotypic impact of a molecular variant identified in a biological sample based on a functional score, a functional classification, an evidence score, or an evidence classification of the identified molecular variant.

[0044] The embodiments herein integrate methods, techniques and technologies from multiple fields. Although statistical and machine learning techniques have been developed and applied to the analysis of molecular measurements of single cells from dozens of (e.g., less than 10 2 However, it is difficult to classify model systems (e.g., cells) of different tissues or developmental stages, but thousands of cells within the same cell line, tissue, or developmental stage have subtle differences (e.g., more than 3×10 9 The requirement to achieve accurate genotype-specific (e.g., molecular variant-specific) classification between cells that lack a single nucleotide difference in a genomic background defined by nucleotides poses considerable challenges.

[0045] The present disclosure provides deep mutational learning (DML) system, apparatus, device, method and / or computer program product embodiments, and / or combinations and sub-combinations thereof, for overcoming challenges in identifying (e.g., classifying) the phenotypic effects of molecular variants identified in a subject based on biological signals measured in a single model system (e.g., cell) and in populations of model systems (e.g., cells).

[0046] The present disclosure provides system, apparatus, device, method and / or computer program product embodiments, and / or combinations and sub-combinations thereof, which provide a method for implementing a DML process and system with a low-cost prediction model by (i) directing deployment of a DML process and system with a low-cost prediction model (see Fig.16 ), and (ii) hierarchical deployment of DML processes and systems that allow robust reconstruction of molecular signals at low cost (see Figure 6 ), to improve cost-efficiency in the classification of molecular variants.

[0047] The present invention provides system, device, apparatus, method and / or computer program product embodiments, and / or combinations and sub-combinations thereof, which improve scalability and performance on each functional element (e.g., gene) by utilizing DML processes and systems for information between functional elements (see Figure 3A and Figure 3B ).

[0048] The present invention provides systems, devices, apparatus, methods and / or computer program product embodiments, and / or combinations and sub-combinations thereof, for evaluating the phenotypic effects (e.g., pathogenicity, functionality or relative effects) of one or more molecular (e.g., genotypic) variants in one or more (e.g., coding or non-coding) functional elements (e.g., protein-coding genes, non-coding genes, molecular domains such as protein or RNA domains, promoters, enhancers, silencers, regulatory binding sites, replication origins, etc.) of a genome (e.g., nuclear, mitochondrial, etc.) or its derived molecules. As will be appreciated by one of ordinary skill in the art, molecular variants may be genotypic (e.g., sequence) variants, such as single nucleotide variants (SNVs), copy number variants (CNVs) or insertions or deletions affecting coding or non-coding sequences (or both) in a nuclear genome, mitochondrial genome, or episomal genome, which may be natural or synthetic. As will be appreciated by one of ordinary skill in the art, molecular variants may also be single amino acid substitutions in protein molecules, single nucleotide substitutions in RNA molecules, single nucleotide substitutions in DNA molecules, or any other molecular alterations of polymeric biomolecules to homologous sequences.

[0049] In some embodiments, classification (or regression) can be based on the molecular variants identified in the biological sample of the subject or its record, involving (e.g., possible) pathogenicity (e.g., pathogenicity) and neutral (e.g., benign) variants of the disorder with a genetic component, or the prediction of its severity. In some other embodiments, classification (or regression) can be based on molecular variants with possible molecular results (e.g., nonsense mutations or insertion mutations and deletion mutations) and possible molecular neutrality (e.g., synonymous), involving molecular effects (e.g., loss of function, gain of function or neutrality). In some other embodiments, classification (or regression) can be based on molecular variants identified in the biological sample of the subject or its record, involving variations in response to treatment (e.g., chemical, biochemical, physical, behavioral, digital or other means). In some embodiments, phenotypic impact can refer to phenotypic categories (e.g., neutral, pathogenic, benign, high risk, low risk, positive response variants, negative response variants) and phenotypic scores (e.g., the possibility of developing specific clinical and non-clinical phenotypes, the level of metabolites in the blood, and the rate at which specific compounds are absorbed or metabolized).

[0050] In some embodiments, the disclosure provides systems and methods for modeling the diversity and universality of phenotypic characteristics within a population based on the diversity and universality of molecular variants in representative populations. In some embodiments, the disclosure provides systems and methods for modeling the diversity and universality of phenotypic characteristics within a population based on the phenotypic effects of molecular variants (with known or expected diversity and universality), wherein the phenotypic effects can be modeled from one or more molecular signals, phenotypic signals, or population signals previously associated with variants in an in vivo or in vitro functional model system. In some embodiments, such modeling can be used to inform the diversity and universality of drug resistance mechanisms in a population.

[0051] In some embodiments, the present disclosure describes the use of models of the diversity and prevalence of phenotypic traits within a population of individuals (e.g., as informed by the phenotypic effects of molecular variants modeled from one or more molecular signals, phenotypic signals, or population signals in a functional model system) to construct cohorts of subjects (e.g., patients), and to study the efficacy of therapeutic and non-therapeutic interventions.

[0052] In some embodiments, the disclosure provides systems and methods for classifying (or regressing) the phenotypic impact of molecular variants based on functional scores or functional classifications derived from one or more molecular signals, phenotypic signals, or population signals associated with the variants determined in a functional model system. In some embodiments, molecular variants can be functionally modeled in cells, cell compartments, or synthetic compartments as in vivo or in vitro model systems.

[0053] In some embodiments, molecular variants modeled (e.g., in vivo or in vitro) can be identified directly in the nucleic acid sequence of the modeled functional element by library preparation, sequencing and characterization of nucleic acids or nucleic acid fragments in a single cell, cell compartment, subcellular compartment or synthetic compartment (e.g., collectively referred to as a model system). In some other embodiments, a pre-assembled database of associated barcodes and variants can be used to infer the molecular variants modeled (e.g., in vivo or in vitro) from the barcode sequence associated with each variant in the functional element by library preparation, sequencing and characterization of nucleic acids or nucleic acid fragments in a model system (e.g., a single cell, cell compartment, subcellular compartment or synthetic compartment). As will be understood by those of ordinary skill in the art, molecular variants can be generated by a variety of techniques such as direct (e.g., chemical) synthesis, error-prone PCR, oligonucleotide-directed mutagenesis, nicking mutagenesis or saturation genome editing (SGE) (Firnberg et al., 2012; Kitzman et al., 2014; Wrenbeck et al., 2016; and Findlay et al., 2014). As will be appreciated by one of ordinary skill in the art, the variant library can then be introduced (e.g., added) into a model system (e.g., a cell, a cellular compartment, a subcellular compartment, or a synthetic compartment) using a variety of methods, such as, but not limited to, homologous recombination (e.g., Cas9-mediated or adenovirus-mediated recombination), site-specific recombination (e.g., Flp-mediated recombination), or viral transduction (e.g., lentivirus-mediated transduction) (Findlay et al., 2018; Wissink et al., 2016; and Macosko et al., 2015).

[0054] In some embodiments, the functional scores and functional classifications associated with each molecular variant can be derived from the measurement of molecules and / or chemical modifiers present in an in vivo or in vitro model system, which contains variants in functional elements (including but not limited to DNA, RNA, and protein molecules or their modifications). For example, in some embodiments, molecular signals, cell signals, or population signals can be measured or modeled and used to learn functional scores and / or functional classifications. In some embodiments, functional scores and functional classifications can be derived from molecular measurements obtained by nucleic acid barcoding, separation, enrichment, library preparation, sequencing, and characterization of multiple nucleic acids or nucleic acid fragments (including but not limited to RNA molecules, genomic DNA, chromatin-associated DNA, protein-associated DNA, accessible DNA fragments, or chemically modified nucleic acids) in a single cell, a cell compartment, a subcellular compartment, or a synthetic compartment. In some embodiments, these processes can use molecular barcoding technology to uniquely identify or associate nucleic acids, nucleic acid fragments, or nucleic acid sequences ( Macosko et al., 2015; Buenrostro et al., 2015; Cusanovich et al., 2015; Dixit et al., 2016; Adamson et al., 2016; Jaitin et al., 2016; Datlinger et al., 2017; Zheng et al., 2017; Cao et al., 2017 These approaches can build on developments in the field of single-cell genomics ( Schwartzman and Tanay 2015; Tanay and Regev 2017; Gawad et al. 2016 Year). In some embodiments, the systems and methods of the present disclosure can apply methods for single-cell RNA sequencing to derive molecular measurements from single cells, cellular compartments, subcellular compartments, or synthetic compartments. These methods include, but are not limited to, single-cell sequencing library generation, high-throughput nucleic acid sequencing, sequencing read quality control, barcode identification (e.g., barcode identification of single cells, cellular compartments, subcellular compartments, or synthetic compartments) and quality control, sequencing read unique molecular barcode identification and quality control, sequencing read alignment, and read alignment filtering and quality control. In some embodiments, the molecular measurements can correspond to site-specific measurements of gene expression (e.g., abundance of RNA transcripts), abundance or modification of proteins (e.g., abundance of phosphorylated proteins), chromatin accessibility (e.g., nucleosome occupancy), epigenetic modifications (e.g., DNA methylation), regulatory activity (e.g., transcription factor binding), post-transcriptional processing (e.g., splicing), post-translational modifications (e.g., ubiquitination), mutation load (e.g., counts), mutation rate (e.g., frequency), mutation signature (e.g., counts or frequencies of each mutation type), or various other types of measurements of molecules within a single cell, cellular compartment, subcellular compartment, or synthetic compartment as will be appreciated by one of ordinary skill in the art. In some embodiments, the present invention describes systems and methods for enhancing the quality of molecular measurements of specific target genes and functional elements by using targeted enrichment or targeted capture techniques (via hybridization-based or amplicon-based techniques and probes) before, during, or after processing of single-cell RNA libraries.

[0055] In some embodiments, molecular measurements from single cells, cellular (or subcellular) compartments, or synthetic compartments can be used to derive multi-site measurements of molecular processes. For example, measurements of these molecular processes can include multi-site measurements of gene expression, chromatin accessibility, epigenetic modifications, regulatory activity, transcriptional activity, translational activity, signaling activity, signaling activity, pathway activity, mutation load, mutation rate, mutation signature, and various other measurements that will be understood by those of ordinary skill in the art.

[0056] In some embodiments, molecular measurements and molecular processes from single cells, cellular (or subcellular) compartments, or synthetic compartments can be used to derive global (e.g., pan-locus or locus-independent) measurements of molecular features. For example, measurements of these molecular features can include global measurements of gene expression, chromatin accessibility, epigenetic modifications, regulatory activity, transcriptional activity, translational activity, signaling activity, signaling activity, pathway activity, mutation load, mutation rate, mutation signatures, and various other measurements that will be understood by those of ordinary skill in the art.

[0057] In some embodiments, molecular measurements, molecular processes, or molecular features of a single cell, cellular compartment, subcellular compartment, or synthetic compartment can be used directly as a (e.g., low-order) molecular score. In some embodiments, a (e.g., high-order) molecular score can be derived by applying a pre-existing model that associates multiple low-order (e.g., low-order) molecular scores (e.g., molecular measurements, molecular processes, or molecular features) to the activity, alteration, defect, or state of regulation, signaling, pathways, processing, cell cycle. In some embodiments, such methods can apply gene set enrichment analysis or other derivative methods that will be understood by those of ordinary skill in the art. In some embodiments, such as Figure 8 As shown in , molecular measurements, molecular processes, molecular features, or (e.g., low-order) molecular scores 806 from single cells, cellular compartments, subcellular compartments, or synthetic compartments having the same molecular variant 802 can be fed through a series of artificial neuron layers (e.g., convolutional layers or perceptron layers) in an artificial neural network 804 (ANN) to derive increasingly complex (e.g., high-order) molecular scores 806 and generate an autoencoder with the learned features. In some embodiments, methods for calculating molecular scores, such as pathway level analysis, can be used to retain information about biological function while allowing dimensionality reduction.

[0058] In some embodiments, such as Fig. 9 As shown in FIG. 1 , a database of molecular scores can be constructed from multiple independent single cells, cellular compartments, subcellular compartments, or synthetic compartments by a cell scoring layer 902. In some embodiments, a database of molecular scores from multiple cells having the same molecular variant 906 (e.g., v 1 、v 2 and v 3) can be accessed with variant sampling layer 908 and analyzed in variant scoring layer 910 to derive (e.g., directly measure or model) summary statistics about the trend (e.g., mean, median, mode), dispersion (e.g., variance, standard deviation), shape (e.g., skewness, kurtosis), probability (e.g., quantile), range (e.g., confidence interval, minimum, maximum), error (e.g., standard error), or covariation (e.g., covariance) associated with each molecular variant. In some embodiments, as Fig. 9 As shown in , summary statistics about the trend, discreteness, shape, range, or error of the molecular scores can be used to create a database of (e.g., quality control) molecular signals 912 associated with each molecular variant 906. In some embodiments, the molecular measurements, molecular processes, molecular features, and molecular scores 904 can be characteristics of each single cell, cellular compartment, subcellular compartment, or synthetic compartment. In some embodiments, the molecular signal can be a characteristic of a molecular variant.

[0059] As will be understood by one of ordinary skill in the art, molecular measurements, processes, features, and scores from a model system (e.g., a single cell, a cellular compartment, a subcellular compartment, or a synthetic compartment) can define or correspond to a unique molecular state or a specific subpopulation of a model system (e.g., a single cell, a cellular compartment, a subcellular compartment, or a synthetic compartment) with similar molecular properties. Fig.10 As shown in FIG. 1 , the cell scoring layer 1002 can be applied to determine the molecular state of the model system, the phenotypic score 1006 (e.g., s 1 、s 2 、s 3 ).

[0060] For example, the molecular state of a model system can be identified based on a cell cycle signature derived from a gene expression molecular score. Macosko et al., 2015 As will be appreciated by one of ordinary skill in the art, scoring can be performed using a previously derived model, for example, by assigning a gene expression signature that has previously characterized a molecular state (such as a gene expression signature previously characterized in chemically synchronized cells reflecting different cell cycle phases ( Whitfield et al., 2002 Year)) to derive the molecular state. As will be appreciated by one of ordinary skill in the art, the molecular state can also be derived by scoring using a partitioned internally derived model from the model system, in which characteristic correlations between molecular signals can be detected or anticipated (e.g., changes in gene expression during different stages of the cell cycle are the case). As will be appreciated by one of ordinary skill in the art, a variety of statistical techniques (e.g., machine learning techniques) can be used to generate the internally derived model.

[0061] In some embodiments, such as Figure 7 As shown in FIG. 1 , the present disclosure provides a method for generating a phenotypic model (m ) for deriving a phenotypic score by using statistical techniques (e.g., machine learning techniques). P ) systems and methods, wherein the statistical techniques relate the molecular scores and molecular states of the model systems (e.g., single cells, cellular compartments, subcellular compartments, or synthetic compartments) to the phenotypic effects of the molecular variants within each model system. Although the molecular scores can be directly related to the molecular properties, biological properties, or physical properties within each model system, the phenotypic scores can describe the (e.g., possible) phenotypic associations of the molecular variants. In some embodiments, the phenotypic scores are derived by applying supervised learning techniques to relate the phenotypic effects (e.g., labels) of the molecular variants within the model system to the molecular scores or molecular states (e.g., features) of the model system.

[0062] In some embodiments, the phenotypic model (m P ) and a database of phenotypic scores (or phenotypic classifications) are generated by accessing a database of features describing (e.g., low-order and high-order) molecular scores and molecular states 704 of single cells 702 and input labels 708 (e.g., a database) describing the phenotypic effects 706 of molecular variants identified within single cells 702. In some embodiments, the training / validation layer 710 generates a phenotypic model (m) that can predict the phenotypic effects 706 of each single cell 702. P ) and quality control thereof. In some embodiments, a feature database describing the molecular scores and molecular states 716 of single cells (tests) 714 is provided to the generated phenotypic model (m P ) to calculate and create a database of phenotypic scores 720 describing the predicted phenotypic effects 718 of molecular variants in single cells (tests) 714. As will be appreciated by one of ordinary skill in the art, the performance (e.g., accuracy) of the predicted phenotypic effects 718 (e.g., phenotypic scores 720) in each cell can be determined against the known phenotypic effects of molecular variants in single cells (tests) 714 within the test layer 712. As will be appreciated by one of ordinary skill in the art, the phenotypic model (m) can be applied to P) to pre-calculate or calculate the phenotypic scores of single cells not included in training, validation or testing as needed. In some embodiments, such scoring and evaluation can be performed in the phenotypic scoring and classification layer 722. The phenotypic scoring and classification layer 722 can be based on the phenotypic score 720 test to allow the phenotype to affect the classification accuracy.

[0063] In some embodiments, summary statistics about the trend, dispersion, shape, range, or error of phenotypic scores can be used to create a database of (eg, quality-controlled) phenotypic signals associated with each molecular variant.

[0064] In some embodiments, such as Fig.10 As shown in , the present disclosure describes the use of molecular state-specific molecular signals in the generation of molecular state-specific models or multi-state models for use in subsequent rounds of unsupervised learning and supervised learning. In some embodiments, as Fig.10 As shown in FIG. 1 , the present disclosure describes the use of a molecular state-specific, variant-specific sampling layer 1008 to access samples having a particular molecular variant 1010 (e.g., v 1 、v 2 、v 3 ) and in a specific molecular state, with a characteristic phenotypic score, or a combination thereof, the molecular measurements, processes, features, and scores 1004 and the molecular state, phenotypic score 1006 of the model system. In some embodiments, the molecular measurements, processes, features, and scores 1004 or the molecular state, phenotypic score 1006 can be pre-computed or calculated as needed by the cell scoring layer 1002. In some embodiments, the data model accessed by the molecular state-specific, variant-specific sampling layer 1008, the summary statistics model, the descriptive statistics (e.g., univariate analysis, bivariate analysis, or multivariate analysis) model, the inference statistics model, the Bayesian inference statistics model (e.g., a variational Bayesian inference model), the Dirichlet process, or other model of the data is used to construct a molecular, phenotypic signal matrix 1012 describing the molecular signals and phenotypic signals of each molecular variant in each molecular state.

[0065] In some embodiments, the molecular, phenotypic signal matrix 1012 can be pre-calculated or calculated as needed. In some embodiments, the molecular, phenotypic signal matrix 1012 can be pre-calculated or calculated as needed by the molecular state, variant-specific scoring layer 1016 to obtain a molecular state-specific matrix. In some embodiments, the molecular, phenotypic signal matrix 1012 can be pre-calculated or calculated as needed by the multi-state, variant-specific scoring layer 1014 to obtain a matrix containing data from multiple molecular states.

[0066] In some embodiments, such as Fig.11As shown in , the present disclosure provides methods for characterizing the distribution of cells with specific molecular variants in terms of molecular states (e.g., subpopulations) or phenotypic scores 1106 (as generated by a cell scoring layer 1102 using molecular measurements, processes, features, and scores 1104 as input). These molecular states (e.g., subpopulations) or phenotypic scores can be associated with, but not limited to, cell subpopulations defined by: (a) characteristic levels of molecular signals (e.g., cyclin-dependent kinases during cell cycle phases) or correlations therebetween, whether determined by applying a pre-existing model or an internally derived model, (b) characteristic levels of phenotypic scores or correlations therebetween, or (c) unsupervised or supervised machine learning methods, including but not limited to dimensionality reduction techniques, examples of which include but are not limited to principal component analysis (PCA), independent component analysis (ICA), and t-stochastic neighbor embedding (tSNE). In some embodiments, as Fig.11 As shown in , for each individual molecular variant 1110, the population sampling layer 1108 generates a measure of the relative representation (e.g., distribution, probability, etc.) of cells at each molecular state (e.g., the proportion or probability of cells with the variant staying in a certain molecular state) or a measure of the relative representation (e.g., distribution, probability, etc.) of cells at each phenotypic score (e.g., the proportion or probability of cells with the variant having a specific score), and can be used to provide a population signal matrix 1112 that describes how the molecular variant affects cells at the population level. The population signal matrix 1112 can contain multiple population signals for multiple molecular variants.

[0067] In some embodiments, subsampling of molecular measurements, molecular processes, molecular features, molecular scores, or phenotypic scores from a model system (e.g., a single cell, a cellular compartment, a subcellular compartment, or a synthetic compartment) with the same molecular variants can be applied to generate independent or disjoint estimates of summary statistics about the trend, discreteness, shape, probability, range, covariation, or error of the molecular measurements, molecular processes, molecular features, or molecular scores or phenotypic scores associated with each molecular variant.

[0068] In some embodiments, independent or disjoint estimates of summary statistics of trends, discreteness, shapes, probabilities, ranges, covariations, or errors for molecular measurements, molecular processes, molecular features, or molecular or phenotypic scores can be used to create a database of (quality-controlled) independent or disjoint estimates of molecular or phenotypic signals associated with each molecular variant. As will be appreciated by one of ordinary skill in the art, independent or disjoint estimates of molecular or phenotypic signals can be used to create a database of (quality-controlled) molecular or phenotypic signals associated with each molecular variant.

[0069] In some embodiments, the present disclosure describes systems and methods for deriving independent or disjoint estimates of summary statistics about trends, discreteness, shapes, probabilities, ranges, covariations, or errors of molecular measurements, molecular processes, molecular features, or molecular scores associated with each molecular variant within a subpopulation of a model system (e.g., a single cell, a cellular compartment, a subcellular compartment, or a synthetic compartment) from a particular molecular state. As will be appreciated by one of ordinary skill in the art, these methods can utilize a variety of statistical techniques (e.g., machine learning techniques).

[0070] In some embodiments, molecular state-specific independent or disjoint estimates of summary statistics regarding trend, discreteness, shape, probability, range, covariation, or error of molecular measurements, molecular processes, molecular features, or molecular or phenotypic scores can be used to create a database of (e.g., quality-controlled) molecular state-specific independent and disjoint estimates of molecular signals or phenotypic signals associated with each molecular variant in a particular molecular state.

[0071] In some embodiments, independent or disjoint estimates of summary statistics about the trend, dispersion, shape, probability, range, covariation, or error of population signals associated with each molecular variant can be used to create a database of (e.g., quality controlled) population signals associated with each molecular variant.

[0072] In some embodiments, such as Fig.12 As shown in , the present disclosure provides systems and methods that utilize a feature extraction layer 1208 (e.g., unsupervised learning techniques) to identify high-order molecular signals, phenotypic signals, or population signals from low-order molecular signals, phenotypic signals, or population signals 1204 associated with each molecular variant 1202, including, but not limited to, feature learning (or representation learning) techniques that deploy an artificial neural network (ANN) 1210 to generate an automatic encoder that can utilize the underlying associations to generate a high-order representation of low-order molecular signals, phenotypic signals, or population signals. In some embodiments, these methods allow for the construction of a database of signals, phenotypic signals, and population signals 1214 for low-order and high-order molecules. In some embodiments, in addition to the low-order molecular signals, phenotypic signals, or population signals 1204, the feature extraction layer 1208 can also access or receive data from the annotation features 1206. In some embodiments, annotation features 1206 may encompass multiple independent (e.g., unmeasured) features that describe variations associated with genotypic variations (e.g., sequence variants, molecular variants, etc.) (e.g., evolutionary features, population features, functional (e.g., annotation-based) features, structural features, kinetic features, and physicochemical features associated with variants, genomic coordinates, transcript (e.g., RNA) coordinates, translation (e.g., protein) coordinates, amino acids, and various other terms that would be understood by one of ordinary skill in the art).

[0073] In some embodiments, the disclosure describes the use of molecular state-specific low-order molecular signals or phenotypic signals to derive molecular state-specific high-order molecular signals or phenotypic signals. In some embodiments, the disclosure describes the use of a multi-state matrix of low-order molecules, phenotypic or population signals to derive multi-state high-order molecules, phenotypic or population signals, which utilizes structured relationships between molecular signals under each molecular state, such as structured gene expression patterns (e.g., molecular signals) under each cell cycle stage (e.g., molecular state). In some embodiments, the disclosure describes the use of convolutional neural networks (CNNs) to learn pattern associations in molecular signals, phenotypic signals, or population signals (and annotation features) under each molecular state.

[0074] In some embodiments, such as Fig.13 As shown in FIG. 1 , the present disclosure provides a method for generating a functional model (m) by statistical (eg, machine) learning. F ) to obtain a system and method for functional scores and functional classification, the functional model (m F ) relate molecular, phenotypic, or population signals (e.g., traits) (single or multiple molecular measurements, molecular processes, molecular signatures, and molecular scores) to the phenotypic effects (e.g., signatures) of molecular variants through regression and classification techniques, respectively.

[0075] In some embodiments, a functional model (m) is generated by accessing a database of features describing molecules (e.g., low-order or high-order), phenotypes, or population signals 1304 of molecular variants 1302 used for training / validation and a set of input labels 1310 (e.g., a database) describing the phenotypic effects 1308 of the molecular variants 1302. F ) and a database of functional scores (or functional classifications). The generating is further performed by applying statistical (e.g., machine) learning techniques to associate molecular, phenotypic, or population signals 1304 (e.g., features) with phenotypic effects (e.g., labels).

[0076] In some embodiments, the training / validation layer 1312 performs training and validation to generate a quality control functional model (m F In some embodiments, the training / validation layer 1312 may deploy cross-validation techniques such as, but not limited to, K-fold cross-validation or leave-one-out cross-validation (LOOCV). In some embodiments, a feature database of molecules, phenotypes, or population signals 1318 describing molecular variants (tests) 1316 may be provided to the generated functional model (m F) to calculate and create a database of functional scores 1324 describing the predicted phenotypic effects 1322 of the molecular variants (tests) 1316. As will be appreciated by one of ordinary skill in the art, the performance (e.g., accuracy) of the predicted phenotypic effects 1322 (e.g., functional scores 1324) of the molecular variants can be determined against the known phenotypic effects of the molecular variants (such as the test molecular variants 1316). As will be appreciated by one of ordinary skill in the art, the functional model (m F ) to pre-calculate or calculate functional scores for molecular variants not included in the training, validation, or testing phases as needed within the testing layer 1314. In some embodiments, such scoring and evaluation can be performed in the functional scoring and classification layer 1326 to, for example, examine the phenotypic impact classification accuracy allowed based on the functional score 1324.

[0077] In some embodiments, the functional model (m F ) during training and testing (prediction generation). In some embodiments, the annotation features 1306 and 1320 may encompass multiple independent (e.g., unmeasured) features describing variations associated with genotypic variations (e.g., sequence variants, molecular variants) (e.g., evolutionary features, population features, functional (e.g., annotation-based) features, structural features, kinetic features, and physicochemical features associated with variants, genomic coordinates, transcript (e.g., RNA) coordinates, translation (e.g., protein) coordinates, amino acids, and various other terms that will be understood by one of ordinary skill in the art).

[0078] As will be appreciated by one of ordinary skill in the art, a variety of sources of phenotypic effects (e.g., signatures) of molecular variants can be used to define a true set, including (e.g., public and / or private) clinical and non-clinical variant databases (e.g., ClinVar, HumVar, VariBench, SwissVar, PhenCode, PharmGKB, or site-specific databases), as well as outcome databases.

[0079] In some other embodiments, the present disclosure provides methods for generating a functional model (m) by statistical (eg, machine) learning. F ) to obtain a system and method for functional scores and functional classification, the functional model (m F) associates a molecule, phenotype, or population signal (e.g., a signature) (derived from one or more molecular measurements, molecular processes, molecular features, and / or molecular scores) with a phenotypic impact (e.g., a signature) of a molecular variant calculated directly from a unique molecule, phenotype, or population signal through regression and classification techniques. In some embodiments, the method may allow, for example, to derive a functional score and functional classification that can predict the relative mutation load, mutation rate, or mutation signature of a sample from a subject with a particular molecular variant. In some embodiments, a functional score or functional classification from such an assay may allow for informing a test subject of the lifetime risk of developing cancer.

[0080] As will be understood by those skilled in the art, the F ) can rely on various statistical (e.g., machine) learning techniques for semi-supervised learning or supervised learning, including but not limited to random forests (RF), gradient boosted trees (GBT), zero rule (ZR), naive Bayes (NB), simple logistic regression (LR), support vector machines (SVM), k nearest neighbors (kNN), and methods of deploying various artificial neural network (ANN) architectures and techniques. In some embodiments, the present disclosure describes the use of molecular state-specific molecular signals to derive molecular state-specific functional scores or functional classifications. In some embodiments, the present disclosure describes the use of a multi-state matrix of molecular signals to derive molecular state-aware functional scores or functional classifications. In some embodiments, the present disclosure describes the use of a convolutional neural network (CNN) to learn the pattern association between functional scores or functional classifications and molecular signals distributed in each molecular state.

[0081] Figure 1AThe application of DML processes and systems in genes of the RAS / MAPK pathway according to some embodiments is shown. The RAS / mitogen-activated protein kinase (MAPK) pathway can play a role in cell proliferation, differentiation, survival and death, and somatic mutations in RAS / MAPK genes can play a role in the occurrence, progression and treatment response of various cancer types through activation and dysregulation of MAPK / ERK signaling. In addition, inherited (e.g., germline) mutations in RAS / MAPK genes have been associated with a variety of autosomal dominant congenital syndromes (including but not limited to Noonan syndrome (NS), Costello syndrome (CS) and cardio-facial-skin (CFC) syndrome and LEOPARD syndrome (LS)), which are present in patients with characteristic facial expressions, heart defects, muscle skin abnormalities and mental retardation, as well as skin, inner ear and genital abnormalities (Aoki et al., 2008). For example, mutations in the non-receptor protein tyrosine phosphatase type 11 (PTPN11) and dual-specificity mitogen-activated protein kinase kinase 1 / 2 genes (MAP2K1, MAP2K2) have been repeatedly found in patients with Noonan syndrome and CFC, with PTPN11 mutations present in up to 50% of patients with Noonan syndrome (Aoki et al., 2008).

[0082] Embodiments can use wild-type, somatic and germline molecular variants of key RAS / MAPK pathway components constructed and overexpressed in HEK293 cells, such as HRAS (e.g., G12V), PTPN11 (e.g., E76K and N308D), and MAP2K2 (e.g., F57C and P128Q). Embodiments can select cells with 1 mg / ml puromycin to ensure expression of exogenously introduced functional elements (e.g., genes), and can verify activation of the RAS / MAPK pathway using enzyme-linked immunosorbent assays (ELISAs) for phospho-ERK protein abundance and total ERK protein abundance (see Figure 5 ). To generate single-cell RNA sequencing data, the implementation scheme can use the 10X Genomics Chromium system to target and capture 500 cells of each molecular variant. Capture and subsequent single-cell library generation can be performed according to the manufacturer's recommendations. The resulting library for each functional element (e.g., gene) can be pooled and sequenced on an Illumina MiniSeq sequencer until the average read per cell for each genotype exceeds 30,000 reads / cell. Single-cell RNA sequencing processing (e.g., single-cell quality control, normalization, transcriptome counting, etc.) can be performed using the 10X Genomics Cell Ranger 2.1.0 pipeline and default settings.

[0083] Figure 1B and Figure 1CProjections of mammalian cells (e.g., HEK293) carrying wild-type and mutant PTPN11 and MAP2K2 (molecular variants associated with germline disorders (F57C, P128Q, and N308D) and molecular variants associated with somatic disorders (E76K)) are shown, according to some embodiments. According to some embodiments, cells can be projected onto a two-dimensional plane derived by t-stochastic neighbor embedding (tSNE) based on molecular scores (e.g., low-order) determined from scaled, normalized unique molecular identifier (UMI) counts of single-cell gene expression. For each gene, tSNE projections based on high-order molecular scores are shown, which are obtained by applying common algorithmic standards widely used in the field (e.g., principal component analysis, PCA) and custom-developed solutions (including cell type-specific, gene-specific, or pathway-specific autoencoders (AEs) trained for robust compressed representation of low-order molecular scores). In some embodiments, the autoencoder can be constructed as a neural network with fully connected layers, a symmetrical number of neurons around the middle layers (e.g., across layers), and rectified linear units (ReLu) for activation. In some embodiments, the autoencoder can be trained using an Adam optimizer and optimized for a mean squared error (MSE) loss function.

[0084] like Figure 1B and Figure 1C As shown in , cell projections from customized cell type and pathway specific autoencoders (AEs) can improve hyperdimensional separation between model systems (e.g., cells) with neutral (e.g., wild-type) molecular variants and disease-associated molecular variants (e.g., N308D, E76K) relative to generalized dimensionality reduction algorithms. The denoising autoencoder (AE) was trained on 8.3 million low-order molecular scores from more than 18,800 genes detected in 3,495 single HEK293 cells with wild-type and mutant RAS / MAPK genes. Training was performed for 30 epochs with a mini-batch size of 10, and noise simulation was performed after randomly reducing the UMI count sampling by 5% between each epoch. The architecture of the fully connected symmetric autoencoder used is in Figure 4 Conventional methods in the field of scaling, normalization, and dimensionality reduction of low-order molecular scores may fail to separate tSNE projections of cells with Noonan syndrome (NS; N308D) molecular variants and wild-type PTPN11, while customized cell type- and pathway-specific autoencoders show robust separation of cells with somatic disorder molecular variants (E76K) and germline disorder molecular variants (N308D) from PTPN11 wild-type cells.

[0085] According to some embodiments, Fig.14A and14B The performance of systems and methods for binomial classification of molecular variants with two distinct phenotypic effects is shown, as assayed in mammalian cells having a disease-associated (e.g., pathogenic) genotype (e.g., sequence) variant (e.g., G12V) and a wild-type (e.g., benign) genotype (e.g., sequence) form of the human HRAS gene, or a third member of the RAS / MAPK pathway encoding the oncoprotein h-Ras (also known as transforming protein p21). A small G protein in the Ras subfamily of the Ras superfamily of small GTPases, h-Ras (upon binding to GTP) can activate RAF family kinases (e.g., c-Raf), leading to cellular activation of the MAPK / ERK pathway.

[0086] Fig.14A Projections 1402 of wild-type and mutant mammalian cells (HEK293) onto a two-dimensional plane derived from their normalized single-cell gene expression measurements by t-stochastic neighbor embedding (tSNE) of the cells are shown. Fig.14A As shown in , low-order molecular scores can be derived from molecular measurements of more than 33,500 genes, with an average of approximately 3500 molecular measurements per cell. Principal component analysis (PCA) can be applied to obtain high-order molecular scores that reduce the dimensionality of the low-order molecular scores. A Gaussian mixture model (GMM) can be applied to assign the projected cells to molecular states 1404 based on the low-order molecular scores obtained from their normalized single-cell gene expression measurements (e.g., UMI counts), thereby defining, for example, N=6 cell subpopulations. Pseudo-disease-associated genotypes and benign genotypes can be obtained by randomly assigning mutant cells and wild-type cells to, for example, k P = 15 disease-associated pseudo-groups and k B =15 benign pseudo-groups were generated. In order to train and test the machine learning functional model (m F ), we can use, for example, an 80 / 20 cross-validation scheme to convert the pseudo-group (k P 1–15, k B 1–15) are divided into training and test sets, resulting in, for example, k for each class label (e.g., disease-associated and benign) 训练 = 12 training genotypes and k 测试 = 3 test genotypes, collectively referred to as the true set. This process can be repeated in each of the f = 5 folds, for example, i = 25 iterations, where in each fold, a pseudo-group (e.g., k P 1–15, k B1–15) are sampled and replaced to retain, for example, 20%, 40%, 60%, 80% or 100% of the cells. In each iteration, sub-group and sampling, the low-order molecular signals and high-order molecular signals of the disease-associated genotype and the benign genotype can be calculated as the average of the low-order molecular scores and the high-order scores, respectively. In each iteration, sub-group and sampling, the population signals of the disease-associated genotype and the benign genotype can be determined as the ratio of cells corresponding to each of, for example, N=6 subpopulations. In each iteration, sub-group and sampling, the machine learning functional model (m F ) can be based on k 训练 The low-order molecular signals, high-order molecular signals, or population signals observed in the data can be used to classify disease-associated genotypes and benign genotypes from the true set. This functional model (m F ) to divide the variants. In each iteration, fold, and sampling, the trained functional model (m F ) can be based on k 测试 The low-order molecular signals, high-order molecular signals or group signals of the pseudo-group are used to predict their class labels (e.g., disease-related or benign). Fig. 14B As shown in , this method can produce robust discrimination between disease-associated genotypes and benign genotypes based on low-order molecular signals, high-order molecular signals, and population signals measured within mutant and wild-type cell populations.

[0087] To evaluate the performance of the DML process and system as a scalable solution for accurate identification of disease-associated (e.g., pathogenic) molecular variants in multi-gene and disorder settings, a unified distributed DML processing pipeline was used to preprocess, scale, normalize, reduce dimensionality, and compute molecular and population signals of, for example, three genes of the RAS / MAPK pathway (HRAS, PTPN11, and MAP2K2). Applying a similar training / testing scheme as described above to evaluate classification accuracy, the DML process can achieve (e.g., median) raw classification accuracy rates 202 of approximately 99.9% and approximately 100% in the analysis of somatic cancer driver molecular variants in HRAS (e.g., G12V) and somatic cancer driver molecular variants in PTPN11 (e.g., E76K), respectively, and (e.g., median) raw classification accuracy rates 204 of approximately 98.5% and approximately 96.1%, respectively, in the analysis of molecular variants forming germline (e.g., genetic) disorders in PTPN11 (e.g., N308D) and molecular variants forming germline (e.g., genetic) disorders in MAP2K2 (e.g., F57C, P128Q), as described above. Figure 2AIn the classification of molecular variants known to cause somatic disorders in HRAS, germline disorders in PTPN11, and germline disorders in MAP2K2, the balance accuracy 206, 208 (e.g., Matthew correlation coefficient, MCC) can be about 99.4%, about 100%, about 95.2%, and about 90.1%, respectively, as shown in FIG. Figure 2B Based on the molecular signals and population signals described herein, in the analysis of disease-associated (e.g., somatic and germline, a combination of both) molecular variants, the raw classification accuracy (e.g., ACC) and the balanced classification accuracy (e.g., MCC) can be about 98.4% and about 95.6%, respectively.

[0088] In some embodiments, the present disclosure provides systems and methods for deriving model system-level (e.g., cell-level) phenotypic scores by applying statistical machine learning models to correlate low-order and high-order molecular scores with known phenotypic effects of variants contained within the model system (e.g., cell). Figure 3A and Figure 3B Shown are the cell-level raw classification accuracies of a machine learning model trained to derive phenotypic scores for cells harboring wild-type and mutant forms of MAP2K2 according to some embodiments.

[0089] exist Figure 3A , the germline bar and the enhancement bar may represent the average classification accuracy of test cells with MAP2K2 germline disorder molecular variants that were excluded from training based on the cell phenotype score, where the training was based only on MAP2K2 neutral molecular variants and germline disorder molecular variants (e.g., germline 302) or included data from PTPN11 germline disorder molecular variants (e.g., enhancement 304). Figure 3B The germline 302 column and the boost 304 column in the Graph represent the average classification accuracy of the tested MAP2K2 germline disorder molecular variants excluded from training, which were determined based on the major cell phenotype scores of cell populations with different cell numbers. Figure 3A As shown in , the germline bars and enhanced bars may correspond to the raw accuracy of classification of the tested molecular variants, where training was based only on MAP2K2 neutral molecular variants and germline disorder molecular variants (e.g., germline) or included data from PTPN11 germline disorder molecular variants (e.g., enhanced).

[0090] Figure 3A and 3BData obtained with a logistic regression (LR) classifier trained to perform binary classification of cells with disease-associated molecular variants and cells with wild-type MAP2K2 based on high-order molecular scores calculated as the first 100 principal components from (e.g., scaled and / or normalized) low-order molecular scores are presented. Cell sets for training and testing can be created by partitioning the molecular variants into training bins and test bins, and then partitioning the cells into corresponding training and test sets based on the molecular variant genotype, such that a specific cell set with a specific disease-associated molecular variant is excluded from training. Thus, classification test performance can be calculated on the entire population of cells with the variant excluded from training. As Figure 3A and Figure 3B As shown in , for molecular variants associated with germline (eg, genetic) disorders in MAP2K2, the average per-cell classification accuracy can be about 80.3%.

[0091] In some embodiments, the present disclosure describes the learning and prediction of phenotypic consequences of molecular variants based on molecular, phenotypic or population signals measured in multiple genes, molecular elements within the same, related or interacting pathways. Figure 3A and Figure 3B As shown in , inclusion of data from PTPN11 molecular variants associated with germline (e.g., genetic) disorders can increase the average cell-by-cell classification accuracy of molecular variants in PTPN11 for each germline disorder from about 80.3% (e.g., germline 302) to about 92.8% (e.g., enhancement 304), demonstrating the ability of the disclosed DML methods and systems to identify and exploit coherent cellular properties to accurately classify the phenotypic effects of molecular variants of multiple functional elements. Figure 3A and Figure 3B As shown in , the increased performance of per-cell classification can result in increased classification of molecular variants based on majority type classification from a population of cells with the molecular variants.

[0092] In some embodiments, the disclosure provides a system and method for deriving a functional score and a functional classification of each functional element (e.g., each gene). In some embodiments, the disclosure provides a method for deriving a functional score and a functional classification of numerous functional elements using consistent molecular signals between molecular variants within a plurality of functional elements. In some embodiments, the disclosure describes a system and method for generating a cell population using a combination of mutagenesis techniques, molecular barcode encoding techniques, molecular cloning techniques, and cell pooling techniques, in which the molecular variants in unique functional elements are uniquely created, barcoded, or both.

[0093] In some embodiments, independent or disjoint estimates of molecular, phenotypic, or population signals (e.g., features) can be used to derive independent or disjoint functional scores and functional classifications by statistical (e.g., machine) learning, which associates molecular signals (e.g., features) with phenotypic effects (e.g., signatures) of molecular variants by regression techniques and classification techniques, respectively.

[0094] In some embodiments, feature weights from statistical (e.g., machine) learning models generated using independent or disjoint estimates of each molecule, phenotype, or population signal are calculated, collected, and used for robust feature selection using techniques that will be understood by one of ordinary skill in the art. In some embodiments, the present disclosure provides methods for deriving functional scores and functional classifications by statistical (e.g., machine) learning, which associates the identified robust molecules, phenotypes, or population signals (e.g., robust features) with the phenotypic effects (e.g., labels) of molecular variants by regression and classification techniques, respectively.

[0095] In some embodiments, the present disclosure describes systems and methods for deriving functional scores and functional classifications from multiple statistical (e.g., machine) learning models that utilize independent or disjoint estimates of molecular signals, applying model selection or model combination (e.g., hybrid) techniques. (Pan et al., 2006) Generated.

[0096] In some embodiments where model selection techniques are applied, models can be compared using a model selection criterion that measures the predictive performance of a model or the probability that a model is the true model, and selection can be applied to maximize the estimate of the selection criterion. As will be appreciated by one of ordinary skill in the art, a variety of model selection criteria can be applied, including, but not limited to, Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Cross Validation (CV), Bootstrap ( Efron, 1983; Efron, 1986; Efron and Tibshirani, 1997 ), or an adaptive model selection criterion computed on the training data or the input test data ( George and Foster, 2000; Shen and Ye, 2002; Shen et al., 2004 ), as exemplified by the test input dependency weight (IDW). The IDW of a candidate model can be defined as the probability that the model gives a correct prediction for a given input or a reasonable measure to quantify the model's predictive performance on the input test data ( Pan et al., 2006 ).

[0097] In some other embodiments of applying the model combination technique, the combined model can be generated by applying an ensemble method by taking an equal or unequal weighted average of the outputs of the individual models ( Ripley, 2008; Hastie et al., 2001For example, ensemble methods may include, but are not limited to, Bayesian model averaging, stacking, bagging, random forest, boosting, ARM, and using performance metrics (e.g., AIC and BIC) as the probability distribution of the training data. (Burnham and Anderson, 2003; Hastie et al., 2001) or computed on the input test data ( Pan et al., 2006 ) weights. In some other embodiments where model combination techniques are applied, an artificial neural network (ANN) architecture may be applied to generate a combined model. In some embodiments, the present disclosure describes systems and methods for deriving function scores and function classifications from multiple statistical (e.g., machine) learning models that are generated using a combination of noise control techniques (e.g., Bootstrap Ensemble with NoiseAlgorithm ( Yuval Raviv, 1996 )) generates independent or disjoint estimates of the molecular signals.

[0098] In some embodiments, the present disclosure describes systems and methods for estimating functional scores and functional classifications of molecular variants by applying statistical (e.g., machine) learning techniques to generate inference models (m I ), the inference model models the relationship between (e.g., assay endpoint) functional scores or functional classifications and multiple dependent (e.g., assayed) features (e.g., molecular, phenotypic, or population signals) or independent (e.g., non-assayed) features (e.g., evolutionary, population, functional (e.g., annotation-based), structural, kinetic, and physicochemical features associated with variants, genomic coordinates, transcript (e.g., RNA) coordinates, translation (e.g., protein) coordinates, amino acids, and various other terms that will be understood by those of ordinary skill in the art). As will be understood by those of ordinary skill in the art, such an inference model (m I ) can allow for the estimation of functional scores and functional classifications of molecular variants with or without explicit use of molecular, phenotypic or population signals, molecular measurements, molecular processes, molecular features or molecular scores. In some embodiments, such methods can allow for the inference of sequence-function graphs describing functional scores and functional classifications of molecular variants for which functional scores and functional classifications are not directly measured. In some embodiments, such as Fig.15As shown in , such systems and methods can allow for the inference of a sequence-function graph 1514 describing a functional score or functional classification for all possible non-synonymous variants in a protein-coding gene using functional scores and functional classifications from a sequence-function graph 1502 (representing a subset of possible non-synonymous variants). In some embodiments, such inference can utilize a score regression layer 1504 that accesses as input an annotation matrix 1506 consisting of annotated features 1508, labels 1510, and functional scores 1512. As will be appreciated by one of ordinary skill in the art, a variety of statistical validation and cross-validation techniques can be applied to monitor and ensure the accuracy of the estimated functional scores and functional classifications.

[0099] In some embodiments, such as Fig.16 As shown in , the present disclosure describes systems and methods for determining the phenotypic impact (e.g., pathogenicity, functionality, or relative effect) of molecular variants through a series of modeling layers that (a) collect or generate existing knowledge or reliable predictions of phenotypic impacts, (b) expand the set of molecular variants with known or predicted phenotypic impacts by performing functional modeling (e.g., performed by a functional modeling engine (FME)) on sampled molecular variants with known, high-confidence predicted, and unknown phenotypic impacts, and (c) further refine the set of molecular variants with known or predicted phenotypic impacts through inference modeling. In combination, these layers can expand (or optimize) the set of molecular variants that can be used for functional models (m F )1607 generates the true set range and reduces (or optimizes) the functional model (m F )1607 generated pair reasoning model (m I ) 1609. In some embodiments, these systems and methods can overcome the limitations of training, validation and testing of functional elements (e.g., genes) and backgrounds with limited availability of molecular variants with known phenotypic effects (e.g., pathogenicity, functionality or relative effects). Such systems and methods thus enable the elucidation of the phenotypic effects of molecular variants of functional elements (e.g., genes) that otherwise have limited data for model generation, and can reduce overall costs.

[0100] In some embodiments, such as Fig.16 As shown in , such systems and methods can combine one or more of the following modeling layers to achieve this goal: (1) predictive model (m P )1603, (2) Sampling model (m S )1605,(3) Functional model(m F )1607, and (4) inference model (m I) 1609. In some embodiments, the present disclosure describes systems and methods for accessing molecular variants with known phenotypic effects (e.g., pathogenic or benign) from existing sources to populate a sequence-function graph 1602 describing the phenotypic effects of molecular variants in genes / functional elements. In some embodiments, a well-characterized prediction model (m P ) 1603 can be used to generate an enhanced sequence-function map 1604 that incorporates the phenotypic impact of molecular variants predicted with high confidence. In some embodiments, the sampling model (m S ) 1605 to generate a set of genotypes (e.g., molecular variants) 1606 that contains (a) a true set obtained by selecting or subsampling molecular variants with known or high-confidence predicted phenotypic effects, and (b) a target set of molecular variants with unknown phenotypic effects.

[0101] In some embodiments, the present disclosure describes the use of statistical (e.g., machine) learning to generate a functional model (m F )1607, the functional model associates molecular, phenotypic or population signals with functional scores and functional classifications learned from molecular variants in a true set (e.g., from genotypes 1606) to predict functional scores and functional classifications of molecular variants in a target set (e.g., from genotypes 1606), thereby generating a sequence-function graph 1608 of functional scores.

[0102] In some embodiments, such as Fig.16 As shown in the figure, the functional model (m F ) 1607 to access enhanced truth sets 1611 and 1612, which include molecular and population signals from multiple functional elements (e.g., genes) in the same, related, or interacting pathways. This capability can enable the system to generate functional models (mF) 1607 for functional elements (e.g., genes) with limited or no availability of molecular variants with known or high-confidence predicted phenotypic effects based on molecular, phenotypic, or population signals from functional elements (e.g., genes) with clear mechanisms of action. Figure 3A and Figure 3B An example thereof is shown.

[0103] In some embodiments, an inference model (mI) 1609 can utilize the phenotypic effects of known molecular variants, high confidence predicted molecular variants, and functionally modeled molecular variants, which models the relationship between the phenotypic effects and multiple dependent (e.g., measured) features (e.g., molecular, phenotypic, or population signals) or independent (e.g., non-measured) features (e.g., evolutionary, population, functional (e.g., annotation-based), structural, kinetic, and physicochemical features associated with variants, genomic coordinates, transcript (e.g., RNA) coordinates, translation (e.g., protein) coordinates, amino acids, and various other terms that will be understood by one of ordinary skill in the art) to generate an enhanced sequence-function graph 1610 of functional scores. As will be understood by one of ordinary skill in the art, such an inference model (mI) 1609 can generate an enhanced sequence-function graph 1610 of functional scores. I )1609 can allow estimation of the phenotypic impact of molecular variants with or without explicit use of molecular, phenotypic, or population signals.

[0104] In some embodiments, the present disclosure describes systems and methods for optimizing the cost efficiency of molecular variant classification by deploying a deep mutation learning (DML) process and system in stages on a true set and a target set (query set) of molecular variants. Some embodiments include, for example Figure 6 , wherein (in cell number, read depth optimization 612) a model system (e.g., cell) comprising true set variants is measured at a high model system (e.g., cell) number and read depth to generate a model for a reduced dimensionality (m DR )614 (such as an automatic encoder (m AE )) and functional model (m F )616 optimized high-quality data. In this first stage, the dimensionality reduction and classification accuracy of the target phenotypic effects of the molecular variants can be optimized to determine a combination of the dimensionality reduction model (614), the functional model (616), and the number of cells and the depth of reads (612) that can ensure robust target performance. In some embodiments, subsampling and noise simulation can be used for training and modeling the performance of the dimensionality reduction model and the functional model. Figure 6 As shown in , certain embodiments include a Phase II production 620 step, in which a model system (e.g., cells) containing a target set of variants and optionally a true set of variants can be assayed in a deployment 622 using a number of cells and / or read depth determined to be robust (e.g., optimal or minimal) when deploying a particular dimensionality reduction model 624 and a functional model 626.

[0105] In some embodiments, the present disclosure describes systems and methods for determining the phenotypic impact (e.g., pathogenicity, function, or relative effect) of a molecular variant identified in a biological sample or record of a subject based on the functional scores and functional classifications determined as described above. In some embodiments, a merged, time-stamped record of the functional scores and functional classifications of a set (e.g., a plurality of unique) of molecular variants can be created, evaluated, validated, selected, and applied to determine the phenotypic impact of the molecular variant identified in a biological sample or record of a subject.

[0106] In some embodiments, the present disclosure describes systems and methods for determining the phenotypic impact (e.g., pathogenicity, functionality, or relative effect) of a molecular variant identified within a biological sample or record of a subject based on a predictor score or predictor classification from a computational predictor generated by applying statistical (e.g., machine) learning methods to utilize functional scores and functional classifications.

[0107] In some embodiments, such as Fig.17 As shown in, the present disclosure describes a method for generating (e.g., low-order) variant interpretation engines (VIEs) by statistical (e.g., machine) learning techniques, which can be gene-specific and condition-specific, and which model the phenotypic effects 1712 of molecular variants based on input labels 1714 and annotation matrices 1706 including their functional scores 1702, 1708 (or functional classifications) and other annotation features 1710, wherein the other annotation features 1710 include features commonly used in computational predictor creation, including but not limited to evolutionary, population, functional (e.g., annotation-based), structural, dynamic, and physicochemical features associated with variants and residues of functional elements. In some embodiments, the training and validation layer 1704 can employ cross-validation techniques 1716 (e.g., K-fold or LOOCV) to train and quality control the VIE, which is then evaluated by a test layer 1718 to derive a predictor score 1720 for molecular variant classification.

[0108] In some embodiments, the present disclosure also describes systems and methods for applying model combination techniques to generate pathway-specific and condition-specific (higher-order) variant interpretation engines (VIEs), which can integrate (lower-order) gene-specific and condition-specific variant interpretation engines (VIEs) from multiple genes in a target pathway of interest. In other embodiments, the present disclosure also describes systems and methods for generating pathway-specific and condition-specific (higher-order) variant interpretation engines (VIEs) by statistical (e.g., machine) learning techniques, which model the phenotypic impact of molecular variants based on their functional scores, functional classifications, and other features commonly used in the creation of computational predictors (including but not limited to evolutionary, population, functional (e.g., annotation-based), structural, kinetic, and physicochemical features associated with variants and residues of functional elements).

[0109] In some embodiments, the present disclosure describes systems and methods for determining the phenotypic impact (e.g., pathogenicity, functionality, or relative effect) of molecular variants identified within a biological sample or record thereof of a subject based on hotspot scores and hotspot classifications from mutation hotspots, which are calculated by applying spatial clustering techniques to identify residue networks with specific phenotypic impacts using the functional scores, functional classifications, and molecular signals associated with molecular variants and residues described and enabled herein.

[0110] In some embodiments, the present disclosure describes systems and methods for deriving a matrix of functional distances between molecular variants or their corresponding residues by: (1) calculating a distance metric between molecular variants projected in an N-dimensional space (1 ≤ N ≤ M) defined by a set of M functional scores, functional classifications, and molecular signals (as described above), where N < M when applying dimensionality reduction techniques to reduce the feature-space of the molecular variants. As will be understood by those of ordinary skill in the art, various dimensionality reduction techniques can be applied, including but not limited to techniques that rely on linear transformations (such as in principal component analysis (PCA)) or techniques that rely on nonlinear transformations (such as in various learning techniques (e.g., t-distributed stochastic neighbor embedding (tSNE) and kernel principal component analysis (kPCA))). As will be understood by those of ordinary skill in the art, various distance metrics can be utilized, including but not limited to Euclidean distance, Manhattan distance (e.g., city block), Mahalanobis distance, or Chebyshev distance, as well as various other distances.

[0111] In some embodiments, the present disclosure describes systems and methods for identifying significantly mutated regions (SMRs) and significantly mutated networks (SMNs) by measuring and scoring the phenotype-associated mutation density (e.g., the number of phenotype-associated variants observed per residue) within spatially neighboring residues of a functional element (e.g., a protein-coding gene) by applying spatial clustering techniques on a variety of spatial distance metrics (including functional distance, sequence distance, structural distance, (co)evolutionary distance, and combinations thereof as described and enabled herein).

[0112] In some embodiments, Fig.18 As shown in , the identification of SMR / SMN can apply a training / validation layer 1804 to identify spatial clusters between phenotypically related or functionally related molecular variants 1806 determined based on the commonality of the functional scores of the molecular variants. In some embodiments, these commonalities can be identified from the functional scores of the molecular variants in the sequence-function graph of the protein-coding gene 1802.

[0113] In some embodiments, such as Fig.18 As shown in , identification of SMRs / SMNs in the training / validation layer 1804 may include a series of steps, including but not limited to: (1) SMR / SMN detection technology 1805, used to identify single residues or residue networks enriched in molecular variants with specific phenotypic associations, as previously described (Araya et al., 2016, U.S. Patent Application 20160378915A1), and (2) SMR / SMN selection technology 1815.

[0114] The SMR / SMN detection technology 1805 may include a series of steps, including but not limited to: (1.1) projecting 1810 the phenotype-associated molecular variants 1806 on functional, sequence, structural or (co-)evolutionary dimensions (or a combination thereof), (1.2) applying spatial clustering technology 1812 (e.g., DBSCAN) to detect clusters of spatially adjacent phenotype-associated variants, and (1.3) measuring the mutation density and score of the phenotype-associated variants for each residue in the cluster.

[0115] The SMN detection technique 1805 may also include the steps indicated in 1814, including but not limited to: (1.4) scoring the mutation density probability by (for example) calculating the (e.g., binomial) probability of obtaining k or more (e.g., greater than or equal to k) observed phenotype-associated variants for each cluster given a per-residue mutation rate within each functional element (e.g., protein-coding gene), (1.5) applying a multiple hypothesis correction (MHC) to the mutation density probabilities of the discovered clusters, and (1.6) calculating the false discovery rate (FDR) of the observed (e.g., original or corrected) mutation density probabilities using a background model of the mutation density probabilities, which background model is obtained by randomizing the positions of the observed phenotype-associated variants within each functional element.

[0116] The training / validation layer 1804 may further perform an SMR / SMN selection technique 1815. The SMR / SMN selection technique may include the following steps: (2.1) defining (e.g., raw or corrected) mutation density probabilities and / or false discovery rates (FDRs) as hotspot scores and applying cutoffs to statistically define hotspot classifications, thereby specifying residues in candidate clusters (e.g., sequence 1816, function 1818, and sequence 1820), (2.2) detecting residues in candidate clusters from multiple unique projections / spaces, (2.3) applying an assignment heuristic to assign residues to each cluster (e.g., selecting the cluster with the largest size (e.g., the cluster with the most residues)), and (2.4) identifying SMR / SMNs as a final set of clusters that meet these criteria. The final set of SMR / SMNs may be derived from multiple unique projections (e.g., sequence 1820, function 1818, or sequence, function (combination) 1822).

[0117] In some embodiments, the present disclosure describes systems and methods for identifying SMR / SMNs by measuring and scoring the phenotype-associated mutation density (e.g., the number of phenotype-associated variants observed per residue) within spatially adjacent residues of a functional element (e.g., a protein-coding gene) using spatial clustering techniques on a variety of spatial distance metrics, wherein the phenotype-associated variants can be defined based on the functional scores and functional classifications described herein. As will be appreciated by one of ordinary skill in the art, these methods can allow for the determination of residue clusters in which variants with well-defined phenotypic effects can occur.

[0118] In some embodiments, the present disclosure describes systems and methods for evaluating independent evidence datasets for explaining molecular variants, such as quantitative (e.g., scoring) or qualitative (classification) evidence from computational predictors (e.g., M-CAP, REVEL, SIFT, and PolyPhen2) as well as gene-specific predictors (e.g., PON-P2), mutation hotspots, and population genomics metrics (e.g., allele frequency-based variant classification). Amendola et al., 2016 ), relative to the accuracy, performance, or robustness of the feature scores and feature classifications described herein.

[0119] In some embodiments, the present disclosure describes systems and methods for calculating evaluation metrics for evaluating the consistency between an evidence dataset and the functional scores and functional classifications described herein, and selecting the best performing evidence dataset for variant interpretation and prioritization based on these evaluation metrics. As will be appreciated by those of ordinary skill in the art, various evaluation metrics can be used to evaluate the consistency of an evidence dataset relative to the functional scores or functional classifications described herein. For quantitative evidence (e.g., scores), these evaluation metrics may include Pearson correlation coefficients, Spearman rank correlations, Kendall correlations, and various other evaluation metrics that should be understood by those of ordinary skill in the art. For qualitative evidence (e.g., classifications), these evaluation metrics may include accuracy, Matthews correlation coefficients, Cohen's kappa coefficients, Youden indexes (e.g., informedness), F measures (e.g., F 1 score), true positive rate (e.g., sensitivity or recall), true negative rate (e.g., specificity), positive predictive value (e.g., precision), negative predictive value, positive likelihood ratio, negative likelihood ratio, and diagnostic odds ratio, as well as various other evaluation indicators that should be understood by ordinary technicians in this field.

[0120] In some embodiments, the present disclosure describes systems and methods that can continuously evaluate, validate, and optimize (e.g., select, remove, or modify) various evidence datasets based on the above-mentioned evaluation indicators, and distribute the best performing (e.g., independent) evidence datasets to client systems via an application programming interface (API) for variant interpretation and prioritization practices, thereby determining the phenotypic impact (e.g., pathogenicity, functionality, or relative effect) of molecular variants identified in a subject's biological sample or its record.

[0121] In some embodiments, the present disclosure describes systems and methods for determining the extent of confirmation bias, reporting bias, or outcome bias present in a dataset of variants, including clinical datasets (e.g., ClinVar, HumVar, VariBench, SwissVar, PhenCode, or site-specific databases), population datasets (e.g., ExAC, GnomAD, and 1000 Genomes), or independent evidence datasets for interpreting molecular variants, such as, but not limited to, computational predictors (e.g., M-CAP, REVEL, SIFT, PolyPhen2, and PON-P2). In some embodiments, the present disclosure describes systems and methods for determining bias based on the expected distribution of functional scores, functional classifications, and molecular signals associated with molecular variants and residues described herein.

[0122] In some embodiments, the present disclosure describes systems and methods for evaluating a target variant dataset by measuring and scoring the differences between the distribution of function scores, function classifications, and molecular signals of molecular variants and residues in the target dataset and the expected distribution of function scores, function classifications, and molecular signals of molecular variants from a reference dataset. In some embodiments, the measurement of inherent bias in the target variant dataset may include a series of steps, including but not limited to: (1) collecting function scores, function classifications, and molecular signals associated with molecular variants in the target dataset and the reference dataset, (2) estimating probability density functions of function scores, function classifications, or molecular signals associated with molecular variants in the reference dataset, (3) estimating probability density functions of function scores, function classifications, or molecular signals associated with molecular variants in the target dataset, and (4) measuring the statistical distance between the probability density functions of the target dataset for function scores, function classifications, or molecular signals and the probability density functions of the reference dataset. In some embodiments, the measurement of inherent bias within a target variant dataset comprises a series of steps, including: (5) sampling variants from a reference dataset (e.g., to match the sample population size of the target dataset), (6) estimating a probability density function of a functional score, functional classification, or molecular signal for the sampled reference dataset in step 5, (7) measuring the statistical distance between the probability density function of the functional score, functional classification, or molecular signal derived from the target dataset and the probability density function of the sampled reference dataset, (8) iterating steps 5-8 to obtain a robust estimate and confidence interval of the statistical distance between the probability density function of the functional score, functional classification, or molecular signal for the target dataset and the reference dataset. In some embodiments, the above-described systems and methods for detection and statistical evaluation of bias allow identification of clinical datasets, population datasets, or evidence datasets in which variants have functional scores, functional classifications, or molecular signals that are different from those expected in the reference dataset.

[0123] In some other embodiments, the present disclosure describes systems and methods for evaluating potential biases within an evidence dataset through a series of steps, including but not limited to: (1) dividing the evidence dataset and the reference dataset into matching quantile sets (e.g., for quantitative evidence scoring) or category sets (e.g., qualitative evidence classification); (2) scoring the variants in each set for multiple properties (e.g., evolutionary characteristics, population characteristics, functional (e.g., annotation-based) characteristics, structural characteristics, kinetic characteristics, and physicochemical characteristics associated with the variants); (3) estimating a probability density function for the score of each property in each set (e.g., the evidence set and the reference set); (4) measuring the statistical distance between the probability density function derived from the evidence set and the probability density function derived from the reference set for each property score; and (5) identifying properties for which the scores between the reference set and the evidence set have statistically significant differences.

[0124] In some embodiments, the present disclosure describes systems and methods that can continuously evaluate and select various evidence datasets based on the above-mentioned bias indicators, and distribute the (e.g., independent) evidence datasets with the least bias to client systems via an application programming interface (API) for use in variant interpretation and prioritization practices to determine the phenotypic impact (e.g., pathogenicity, functionality, or relative effect) of molecular variants identified in a subject's biological sample or its record.

[0125] In some embodiments, the present disclosure describes systems and methods for determining the phenotypic impact (e.g., pathogenicity, function, or relative effect) of molecular variants identified within a biological sample or record of a subject based on the functional scores, functional classifications, predictor scores, predictor classifications, hotspot scores, and hotspot classifications described herein in functional elements (e.g., genes) and pathways that are associated with Mendelian diseases (e.g., Table 1), are known cancer driver genes (e.g., Table 2), are pharmacogenomics genes in which genotype (e.g., sequence) variation is associated with drug response variation (Table 3), or are other clinically valuable genes (e.g., Table 4).

[0126] In some embodiments, the present disclosure describes systems and methods for evaluating, selecting, distributing, and utilizing independent evidence (determined to be the best performing and least biased based on the functional scores and classifications described herein) for interpretation and prioritization of variants in functional elements (e.g., genes) and pathways that are associated with Mendelian diseases (e.g., Table 1), are known cancer driver genes (e.g., Table 2), are pharmacogenomics genes in which genotype (e.g., sequence) variation is associated with drug response variation (Table 3), or are other clinically valuable genes (e.g., Table 4).

[0127] As discussed above, Table 1 is an example table of functional elements and pathways associated with Mendelian diseases according to some embodiments. Table 2 is an example table of functional elements and pathways that are known cancer driver genes according to some embodiments. Table 3 is an example table of pharmacogenomics genes in which genotype (e.g., sequence) variation is associated with drug response variation according to some embodiments. Table 4 is an example table of other clinically valuable genes according to some embodiments. Tables 1-4 can be found on page 47 of the specification.

[0128] In some embodiments, the present disclosure describes systems and methods for determining the phenotypic impact (e.g., pathogenicity, function, or relative effect) of a molecular variant identified in a biological sample or record of a subject based on the functional scores, functional classifications, predictor scores, predictor classifications described and enabled herein for variants within known pathogenic variant targets (including, but not limited to, mutation hotspots), or within, for example, 50, 100, 500, and 1,000 base pairs (bp) of such hotspots. In some embodiments, the present disclosure describes systems and methods for determining the phenotypic impact (e.g., pathogenicity, function, or relative effect) of a molecular variant identified in a biological sample or record of a subject based on the functional scores, functional classifications, predictor scores, or predictor classifications of variants within, for example, 50, 100, 500, and 1000 bp of such regions. As will be appreciated by one of ordinary skill in the art, a variety of methods can be applied to determine mutation hotspots and variation-restricted regions.

[0129] For example, one or more computer systems (such as Fig.19 The computer system 1900 shown in FIG. 1 can be used to implement various embodiments. The computer system 1900 can be used, for example, to implement Figure 1A , Figure 6-13 and Figure 15-18 Computer system 1900 may be any computer capable of performing the functions described herein.

[0130] Computer system 1900 may be any known computer capable of performing the functions described herein.

[0131] Computer system 1900 includes one or more processors (also called central processing units or CPUs), such as processor 1904. Processor 1904 is connected to a communication infrastructure or bus 1906.

[0132] One or more processors 1904 may each be a graphics processing unit (GPU). In one embodiment, a GPU is a processor that is a special-purpose electronic circuit designed to process mathematically intensive applications. A GPU may have a parallel structure that can be effectively used for parallel processing of large blocks of data (such as mathematically intensive data common in computer graphics applications, images, videos, etc.).

[0133] The computer system 1900 also includes user input / output devices 1903 , such as a monitor, keyboard, pointing device, etc., which communicate with the communication infrastructure 1906 through the user input / output interface 1902 .

[0134] The computer system 1900 also includes a main memory 1908, such as a random access memory (RAM). The main memory 1908 may include one or more levels of cache. The main memory 1908 stores control logic (eg, computer software) and / or data therein.

[0135] The computer system 1900 may also include one or more secondary storage devices or memories 1910. The secondary storage 1910 may include, for example, a local, network, or cloud accessible hard drive 1912 and / or a removable storage device or drive 1914. The removable storage drive 1914 may be a floppy disk drive, a tape drive, an optical drive, an optical storage device, a tape backup device, and / or any other storage device / drive.

[0136] Removable storage drive 1914 may interact with removable storage unit 1918. Removable storage unit 1918 includes a computer usable or readable storage device having computer software (control logic) and / or data stored thereon. Removable storage unit 1918 may be a floppy disk, magnetic tape, optical disk, DVD, optical storage disk, and / or any other computer data storage device. Removable storage drive 1914 reads and / or writes to removable storage unit 1918 in a well-known manner.

[0137] According to an exemplary embodiment, secondary storage 1910 may include other devices, mechanisms, or other methods for allowing computer system 1900 to access computer programs and / or other instructions and / or data. Such devices, mechanisms, or other methods may include, for example, a removable storage unit 1922 and an interface 1920. Examples of removable storage unit 1922 and interface 1920 may include a program cartridge and cartridge interface (such as those found in video game devices), a removable memory chip (such as an EPROM or PROM) and an associated socket, a memory stick and a USB port, a memory card and an associated memory card slot, and / or any other removable storage unit and associated interface.

[0138] The computer system 1900 may also include a communication or network interface 1924. The communication interface 1924 enables the computer system 1900 to communicate and interact with any combination of remote devices, remote networks, remote entities, etc. (individually and collectively referred to by reference numeral 1928). For example, the communication interface 1924 may allow the computer system 1900 to communicate with a remote device 1928 via a communication path 1926, which may be wired and / or wireless and may include any combination of a LAN, a WAN, the Internet, etc. Control logic and / or data may be transmitted to and from the computer system 1900 via the communication path 1926.

[0139] In one embodiment, a tangible device or article of manufacture including a tangible computer usable or readable medium having control logic (software) stored thereon is also referred to herein as a computer program product or program storage device. This includes, but is not limited to, computer system 1900, main memory 1908, auxiliary memory 1910, and removable storage units 1918 and 1922, as well as tangible articles of manufacture embodying any combination of the foregoing. When executed by one or more data processing devices (such as computer system 1900), such control logic causes such data processing devices to operate as described herein.

[0140] Based on the teachings contained in this disclosure, it will be apparent to those skilled in the relevant art how to use the Fig.12 Embodiments of the present disclosure may be made and used with data processing devices, computer systems, and / or computer architectures other than those shown in . Specifically, embodiments may be operated using software, hardware, and / or operating system implementations other than those described herein.

[0141] It should be understood that the Detailed Description section, and not any other section, is intended to be used to interpret the claims. The other sections may set forth one or more but not all exemplary embodiments contemplated by the inventors and, therefore, are not intended to limit the present disclosure or the appended claims in any way.

[0142] Although the present invention describes exemplary embodiments for exemplary fields and applications, it should be understood that the present invention is not limited thereto. Other embodiments and modifications thereof are possible and are within the scope and spirit of the present disclosure. For example, and without limiting the generality of this paragraph, the embodiments are not limited to the software, hardware, firmware and / or entities shown in the drawings and / or described herein. In addition, the embodiments (whether or not explicitly described herein) have significant utility for fields and applications beyond the examples described herein.

[0143] Embodiments have been described herein with the aid of functional building blocks that illustrate the implementation of specific functions and their relationships. For ease of description, the boundaries of these functional building blocks have been arbitrarily defined herein. Alternative boundaries may be defined as long as the specified functions and relationships (or their equivalents) are properly performed. Moreover, alternative embodiments may perform functional blocks, steps, operations, methods, etc., in an order different from that described herein.

[0144] Reference herein to "one embodiment", "embodiment", "exemplary embodiment" or similar phrases indicates that the described embodiment may include specific features, structures or characteristics, but each embodiment may not necessarily include the specific features, structures or characteristics. In addition, these phrases do not necessarily refer to the same embodiment. In addition, when describing specific features, structures or characteristics in conjunction with an embodiment, it will be within the knowledge of a person skilled in the art to incorporate such features, structures or characteristics into other embodiments, regardless of whether or not explicitly mentioned or described in this article. In addition, some embodiments may be described using the expressions "coupled" and "connected" and their derivatives. These terms are not necessarily synonymous with each other. For example, some embodiments may be described using the terms "connected" and / or "coupled" to indicate that two or more elements are in direct physical or electrical contact with each other. However, the term "coupled" may also mean that two or more elements are not in direct contact with each other, but still cooperate or interact with each other.

[0145] The breadth and scope of the present disclosure should not be limited by any of the above-described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents.

[0146] [Table 1: Mendelian disease_gene (HGNC symbol)]

[0147]

[0148] [Table 2: Cancer driver gene (CCG La)_Gene (HGNC symbol)]

[0149]

[0150]

[0151]

[0152] [Table 3: Pharmacogenomics (Pharm)_Gene (HGNC symbol)]

[0153]

[0154]

[0155]

[0156]

[0157]

[0158] [Table 4: Clinical test gene_gene (HGNC symbol)]

[0159]

[0160]

[0161]

[0162]

[0163]

[0164]

[0165]

References

[0166] Aoki et al., “The RAS / MAPK Syndromes: Novel Roles of the RAS Pathway in Human Genetic Disorders,” Human Mutation, 2008.

[0167] KARCZEWSKI et al., “Analysis of protein-coding genetic variation in 60,706 humans,” Nature, 2016.

[0168] LANDRUM et al., “ClinVar: public archive of interpretations of clinically relevant variants,” Nucleic Acids Res., 2015.

[0169] MAXWELL et al., “Evaluation of ACMG-Guideline-Based Variant Classification of Cancer Susceptibility and Non-Cancer-Associated Genes in Families Affected by Breast Cancer,” Am. J. Hum. Genet., 2016.

[0170] MYERS et al., “The lipid phosphatase activity of PTEN is critical for its tumor supressor function,” Proc. Natl. Acad. Sci. U.S.A., 1998.

[0171] Myers et al., "PTEN, the tumor suppressor from human chromosome 10q23, is a dual-specificity phosphatase," Proc. Natl. Acad. Sci. U.S.A., 1997.

[0172] He et al., "Cowden syndrome-related mutations in PTEN associate with enhanced proteasome activity," Cancer Res., 2013.

[0173] Heikkinen et al., "Variants on the promoter region of PTEN affect breast cancer progression and patient survival," Breast Cancer Res., 2011.

[0174] Johnston et al., "Conformational stability and catalytic activity of PTEN variants linked to cancers and autism spectrum disorders," Biochemistry, 2015.

[0175] Markkanen et al., "DNA Damage and Repair in Schizophrenia and Autism: Implications for Cancer Comorbidity and Beyond," Int. J. Mol. Sci., 2016.

[0176] Scharner et al., "Genotype–phenotype correlations in laminopathies: how does fate translate?," Biochem. Soc. Trans., 2010.

[0177] Araya et al., “Deep mutational scanning: assessing protein function on a massive scale,” Trends Biotechnol., 2011.

[0178] Shendure et al., “Massively Parallel Genetics,” Genetics, 2016.

[0179] Kelsic et al., “RNA Structural Determinants of Optimal Codons Revealed by MAGE-Seq,” Cell Syst, 2016.

[0180] Patwardhan et al., “High-resolution analysis of DNA regulatory elements by synthetic saturation mutagenesis,” Nat. Biotechnol., 2009.

[0181] Buenrostro et al., “Quantitative analysis of RNA-protein interactions on a massively parallel array reveals biophysical and evolutionary landscapes,” Nat. Biotechnol., 2014.

[0182] Guenther et al., “Hidden specificity in an apparently nonspecific RNA-binding protein,” Nature, 2013.

[0183] Araya et al., “A fundamental protein property, thermodynamic stability, revealed solely from large-scale measurements of protein function,” Proc. Natl. Acad. Sci. U.S.A., 2012.

[0184] FOWLER et al., “High-resolution mapping of protein sequence-function relationships,” Nat. Methods, 2010.

[0185] MAJITHIA et al., “Prospective functional classification of all possible missense variants in PPARG,” Nat. Genet., 2016.

[0186] STARITA et al., “Massively Parallel Functional Analysis of BRCA1 RING Domain Variants,” Genetics, 2015.

[0187] BUENROSTRO et al., “Single-cell chromatin accessibility reveals principles of regulatory variation,” Nature, 2015.

[0188] CUSANOVICH et al., “Multiplex single-cell profiling of chromatin accessibility by combinatorial cellular indexing,” Science, 2015.

[0189] CAO et al., “Comprehensive single cell transcriptional profiling of a multicellular organism by combinatorial indexing,” bioRxiv, 2017.

[0190] ZHENG et al., “Massively parallel digital transcriptional profiling of single cells,” Nat. Commun., 2017.

[0191] DATLINGER et al., “Pooled CRISPR screening with single-cell transcriptomereadout,” Nat. Methods, 2017.

[0192] JAITIN et al., “Dissecting Immune Circuits by Linking CRISPR-PooledScreens with Single-Cell RNA-Seq,” Cell, 2016.

[0193] ADAMSON et al., “A Multiplexed Single-Cell CRISPR Screening PlatformEnables Systematic Dissection of the Unfolded Protein Response,” Cell, 2016.

[0194] DIXIT et al., “Perturb-Seq: Dissecting Molecular Circuits with ScalableSingle-Cell RNA Profiling of Pooled Genetic Screens,” Cell, 2016.

[0195] MACOSKO et al., “Highly Parallel Genome-wide Expression Profiling ofIndividual Cells Using Nanoliter Droplets,” Cell, 2015.

[0196] GAWAD et al., “Single-cell genome sequencing: current state of thescience,” Nat. Rev. Genet., 2016.

[0197] TANAY et al., “Scaling single-cell genomics from phenomenology tomechanism,” Nature, 2017.

[0198] SCHWARTZMAN et al., “Single-cell epigenomics: techniques and emerging applications,” Nat. Rev. Genet., 2015.

[0199] BUZDIN et al., “The OncoFinder algorithm for minimizing the errors introduced by the high-throughput methods of transcriptome analysis,” Front Mol Biosci, 2014.

[0200] MACOSKO et al., “Highly Parallel Genome-wide Expression Profiling of Individual Cells Using Nanoliter Droplets,” Cell, 2015.

[0201] WHITFIELD et al., “Identification of genes periodically expressed in the human cell cycle and their expression in tumors,” Mol. Biol. Cell, 2002.

[0202] PAN et al., “Using input dependent weights for model combination and model selection with multiple sources of data,” Stat. Sin., 2006.

[0203] EFRON et al., “Improvements on Cross-Validation: The 632+ Bootstrap Method,” J. Am. Stat. Assoc., 1997.

[0204] EFRON, “How Biased is the Apparent Error Rate of a Prediction Rule?,” J. Am. Stat. Assoc., 1986.

[0205] Efron, “Estimating the Error Rate of a Prediction Rule: Improvement on Cross-Validation,” J. Am. Stat. Assoc., 1983.

[0206] Shen et al., “Adaptive Model Selection and Assessment for Exponential Family Distributions,” Technometrics, 2004.

[0207] Shen et al., “Adaptive Model Selection,” J. Am. Stat. Assoc., 2002.

[0208] George et al., “Calibration and Empirical Bayes Variable Selection,” Biometrika, 2000.

[0209] Ripley et al., “Pattern Recognition and Neural Networks,” Cambridge University Press, 2008.

[0210] Hastie et al., “The Elements of Statistical Learning. Data Mining, Inference, and Prediction,” Springer, 2001.

[0211] Burnham et al., “Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach,” Springer, 2003.

[0212] Yuval, “Bootstrapping with Noise: An Effective Regularization Technique,” Connection Science, 1996.

[0213] AMENDOLA et al., “Performance of ACMG-AMP Variant-InterpretationGuidelines among Nine Laboratories in the Clinical Sequencing Exploratory Research Consortium,” Am.J.Hum.Genet., 2016.

[0214] BERGER et al., “High-throughput Phenotyping of Lung Cancer Somatic Mutations,” Cancer Cell, 2016 30(2); pp. 214–228.

[0215] MACOSKO et al., “Highly Parallel Genome-wide Expression Profiling of Individual Cells Using Nanoliter Droplets,” Cell, 2015 161(5); pp. 1202–1214.

[0216] STARITA et al., “Deep Mutation Scanning: A Highly Parallel Method toMeasure the Effects of Mutation on Protein Function,” Cold Spring Harb Protoc, 2015(8); pp. 711–714.

[0217] SHENDURE et al., “A framework for determining the relative effect of genetic variants,”

[0218] U.S. Patent No. 15 / 023,355, filed March 18, 2016.

[0219] REGEV et al., “A droplet-based method and apparatus for composite single-cell nucleic acid analysis,” International Patent Publication No. WO 2016 / 040476, published March 17, 2016.

[0220] KALIA SS et al., "Recommendations for reporting of secondary findings inclinical exome and genome sequencing,2016update(ACMG SF v2.0):apolicystatement of the American College of Medical Genetics and Genomics,"GenetMed.,2016.

[0221] FUTREAL AP et al., "A census of human cancer genes,"Nat Rev Cancer,2004 4(3); pp. 177–183.

[0222] LAWRENCE MS et al., "Discovery and saturation analysis of cancer genesacross 21tumour types,"Nature,2014 505(7484); pp. 495–501.

[0223] WHIRL-CARRILLO et al., "Pharmacogenomics knowledge for personalizedmedicine,"Clin Pharmacol Ther,2012 92(4); pp. 414–417.

[0224] RUBINSTEIN et al., "The NIH genetic testing registry:a new,centralizeddatabase of genetic tests to enable access to comprehensive information andimprove transparency,"Nucleic Acids Res,2013 4; pp. D925–35.

[0225] SAMOCHA KE et al., (2017) "Regional missense constraint improves variantdeleteriousness prediction,"bioRxiv:148353.

[0226] Kitzman, JO, Starita, LM, Lo, RS, Fields, S. & Shendure, J. Massively parallel single-amino-acid mutagenesis. Nat. Methods 12, 203–206 (2015).

[0227] Findlay, GM, Boyle, Ea, Hause, RJ, Klein, JC and Shendure, J. (2014). Saturation editing of genomic regions by multiplex homology-directed repair. Nature 513, 1–2.

[0228] Firnberg, E. & Ostermeier, M. PFunkel: Efficient, Expansive, User-Defined Mutagenesis. PLoS One 7, 1–10 (2012).

[0229] Wrenbeck, EE et al., Plasmid-based one-pot saturationmutagenesis. Nat. Methods 13, 928–930 (2016).

[0230] Wissink, EM, Fogarty, EA & Grimson, A. High-throughput discovery of post-transcriptional cis-regulatory elements. BMC Genomics 17, 1–14 (2016).

[0231] Araya et al., 2016, U.S. Patent Application No. 20160378915A1.

[0232] This manual also includes the following contents:

[0233] 1. A computer-implemented method for determining the phenotypic impact of a molecular variant identified in a biological sample, comprising:

[0234] (a) receiving a molecular variant associated with one or more functional elements within a model system, wherein the model system comprises a single cell, a cellular compartment, a subcellular compartment, or a synthetic compartment;

[0235] (b) determining a molecular score or a phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment,

[0236] wherein the molecular score is a molecular measurement, a molecular process or a molecular signature, or a combination thereof, of a single cell, a cellular compartment, a subcellular compartment or a synthetic compartment, and

[0237] wherein the phenotypic score is derived by applying supervised learning techniques to relate the phenotypic effects of molecular variants within a model system to the molecular score of the model system;

[0238] (c) determining a molecular signature or a phenotypic signature associated with the molecular variant based on the respective molecular scores or phenotypic scores of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment carrying the molecular variant;

[0239] (d) determining a population signal associated with the molecular variant based on the molecular score or phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment carrying the molecular variant,

[0240] wherein for each individual molecular variant, the population sampling layer produces a measure of the relative representation of cells in each molecular state or a measure of the relative representation of cells in a phenotypic score and is used to provide a population signaling matrix that describes how the molecular variant affects cells at the population level, and

[0241] wherein the population signal matrix contains a plurality of population signals for a plurality of molecular variants;

[0242] (e) determining a functional score or a functional classification of the molecular variant based on statistical learning, wherein the statistical learning correlates the molecular signal, the phenotypic signal or the population signal of the molecular variant with the phenotypic impact of the molecular variant;

[0243] (f) deriving an evidence score or evidence classification for the molecular variant based on the functional score or functional classification, modeling of the functional score or functional classification, modeling of the predictor score or predictor classification, or modeling of the hotspot score or hotspot classification,

[0244] The training and validation layers use cross-validation techniques for training and quality control, which are then evaluated by the test layer to derive a predictor score for molecular variant classification, and

[0245] wherein a mutation density probability and / or a false discovery rate is defined as a hotspot score and a cutoff is applied to statistically define hotspot classifications to designate residues in candidate clusters; and

[0246] (g) determining the phenotypic impact of the molecular variant based on the functional score, the functional classification, the evidence score or the evidence classification.

[0247] 2. The method according to embodiment 1, wherein the evidence score or the evidence classification:

[0248] (i) is determined based on the molecular signal, the phenotypic signal or the population signal from the molecular variant in one or more functional elements;

[0249] (ii) derived from the functional score or functional classification, the predictor score or predictor classification, or the hotspot score or hotspot classification;

[0250] (iii) deriving the evidence score or evidence classification by applying the statistical learning to relate the evidence score or evidence classification to the phenotypic effect of the molecular variant using regression or classification; or

[0251] (iv)A combination of the above.

[0252] 3. A method according to embodiment 1, wherein the functional score or functional classification of the molecular variant is determined by applying statistical learning to associate the molecular signal with the phenotypic impact of the molecular variant using regression or classification.

[0253] 4. The method according to embodiment 1, wherein

[0254] (i) determining the phenotypic impact of the molecular variant based on a clinical database, a phenotypic database, a population database, a molecular annotation database or a functional database of the molecular variant, a subject or a population; and / or

[0255] (ii) determining said phenotypic impact of said molecular variant based on molecular signals such as mutation load, mutation rate and mutation signature.

[0256] 5. The method according to embodiment 1, wherein

[0257] (i) deriving the functional score or functional classification of the molecular variant from a plurality of statistical models generated using independent or disjoint estimates of the molecular signal, the phenotypic signal or the population signal; and / or

[0258] (ii) deriving said functional score or functional classification of said molecular variant from a functional modeling engine (FME),

[0259] wherein the FME is generated by applying machine learning techniques to associate unmeasured features of the molecular variants with the functional scores or functional classifications, and

[0260] The undetermined characteristics include evolutionary characteristics, population characteristics, functional characteristics, structural characteristics, kinetic characteristics and physicochemical characteristics.

[0261] 6. A method according to embodiment 5, wherein the predictor score or predictor classification of the molecular variant is derived from a variant interpretation engine (VIE), wherein the VIE is generated by applying machine learning techniques to associate the functional score or functional classification and the unmeasured features with the phenotypic impact of the molecular variant.

[0262] 7. The method according to embodiment 1, wherein

[0263] (i) the molecular signal comprises a low-order molecular signal of the molecular variant, the low-order molecular signal being derived as a summary statistical model, a descriptive statistical model, an inferential statistical model or a Bayesian inference model of the molecular score measured in the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment carrying the molecular variant;

[0264] (ii) the molecular signature comprises a higher-order molecular signature of the molecular variant, the higher-order molecular signature being derived by applying a pre-existing model that relates lower-order molecular signatures to activity, alteration, defect or state of regulation, signaling, pathway, processing or cell cycle;

[0265] (iii) the molecular signal comprises a high-order molecular signal of the molecular variant, wherein the high-order molecular signal is derived from a low-order molecular signal via an unsupervised learning technique, a feature learning technique or a dimensionality reduction technique;

[0266] (iv) the molecular signal comprises a low-order molecular score corresponding to a molecular measurement result, a molecular process, or a molecular feature from the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment;

[0267] (v) the molecular signal comprises a high-order molecular score for the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment, the high-order molecular score being derived by applying a pre-existing model that relates low-order molecular scores to activity, alteration, defect or state of regulation, signaling, pathway, processing or cell cycle;

[0268] (vi) the molecular signal comprises a high-order molecular score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment, wherein the high-order molecular score is derived from a low-order molecular score via an unsupervised learning technique, a feature learning technique or a dimensionality reduction technique; or

[0269] (vii)A combination of the above.

[0270] 8. The method according to embodiment 7, wherein

[0271] (i) the molecular measurements correspond to site-specific measurements of gene expression, protein expression, chromatin accessibility, epigenetic modification, regulatory activity, post-transcriptional processing, post-translational modification, mutational status, mutational load or mutation rate of a molecule within the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment;

[0272] (ii) the molecular measurements correspond to multi-locus measurements of gene expression, protein expression, chromatin accessibility, epigenetic modification, regulatory activity, transcriptional activity, translational activity, signaling activity, pathway activity, mutational status, mutational load, or mutation rate derived from molecular measurements within the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment;

[0273] (iii) the molecular measurement corresponds to a global measure of gene expression, protein expression, chromatin accessibility, epigenetic modification, regulatory activity, transcriptional activity, translational activity, signaling activity, pathway activity, mutational status, mutational load, or mutation rate derived from a molecular measurement or molecular process within the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment; or

[0274] (iv) any combination of the above, and

[0275] The molecular measurement results may include: sequencing read quality control, cellular barcode identification or quality control, molecular barcode identification or quality control, sequencing read alignment with a reference genome, sequencing read alignment filtering or quality control, mapping filtered and quality controlled sequencing reads to functional elements, mapping filtered and quality controlled molecular barcodes to functional elements, mapping filtered and quality controlled sequencing reads or molecular barcodes of cellular barcodes to functional elements, or a combination of the above.

[0276] 9. The method according to embodiment 1, wherein

[0277] (i) the molecular signal, the phenotypic signal or the population signal is molecular state specific, which is derived from the molecular state of the single cell, the cellular compartment, the subcellular compartment or the population of the synthetic compartment to allow learning in a state-specific learning layer;

[0278] (ii) the molecular signal, the phenotypic signal or the population signal is molecular state agnostic, which is derived from a population of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment in multiple molecular states to allow learning in a state-agnostic learning layer;

[0279] (iii) the molecular signal, the phenotypic signal or the population signal is molecular state ordered, which originates from a population of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment in multiple molecular states to allow learning in a multi-state learning layer; or

[0280] (iv)A combination of the above.

[0281] 10. The method according to embodiment 1, wherein

[0282] (i) deriving the molecular state of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment by applying a pre-existing model that relates a molecular score or a phenotypic score to the molecular state, wherein the pre-existing model assigns single cells to cell cycle phases based on previously characterized gene expression signatures; and / or

[0283] (ii) Determining the molecular state of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment by applying unsupervised learning technology, feature learning technology or dimensionality reduction technology to the molecular score or phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment.

[0284] 11. A method according to embodiment 1, wherein the molecular variant corresponds to a coding or non-coding variant or a previously identified mutation hotspot, wherein the coding or non-coding variant or the previously identified mutation hotspot is within a functional element, gene and pathway associated with (i) Mendelian disease, (ii) a known cancer driver gene, (iii) a drug response variant or (iv) other genes of clinical value.

[0285] 12. A method according to embodiment 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 10bp, 50bp, 100bp, 500bp or 1000bp of previously identified mutation hotspots of functional elements, genes and pathways associated with Mendelian diseases, known cancer driver genes, drug response variants or other genes of clinical value.

[0286] 13. The method according to embodiment 1, wherein

[0287] (i) the phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment comprises a low-order phenotypic score,

[0288] wherein the low-order phenotypic score corresponds to a score or classification generated by the phenotypic model using a statistical learning technique that relates the molecular scores and molecular states of the model systems to the phenotypic effects of the molecular variants within each model system, and

[0289] wherein the database of phenotypic models and phenotypic scores is generated by accessing a database of features describing molecular scores and molecular states of single cells and input labels describing the phenotypic effects of molecular variants identified within the single cells; and / or

[0290] (ii) the phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment comprises a high-order phenotypic score, wherein:

[0291] (a) the higher-order phenotypic score is derived by applying a pre-existing model that relates lower-order phenotypic scores to activity, alteration, defect or state of regulation, signaling, pathway, processing or cell cycle, and / or

[0292] (b) The high-level phenotypic scores are derived from low-level phenotypic scores via unsupervised learning techniques, feature learning techniques or dimensionality reduction techniques.

[0293] 14. The method according to embodiment 1, wherein

[0294] (i) the phenotypic signal associated with the molecular variant comprises a low-order phenotypic signal associated with the molecular variant, wherein the low-order phenotypic signal associated with the molecular variant is derived as a summary statistical model, a descriptive statistical model, an inferential statistical model or a Bayesian inference model of the phenotypic score measured in the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment carrying the molecular variant; and / or

[0295] (ii) the phenotypic signal associated with the molecular variant comprises a high-order phenotypic signal associated with the molecular variant, wherein

[0296] (a) the higher-order phenotypic signal associated with the molecular variant is derived by applying a pre-existing model that relates lower-order phenotypic signals to activity, alteration, defect or state of regulation, signaling, pathways, processing or cell cycle, or

[0297] (b) The high-order phenotypic signal associated with the molecular variant is derived from the low-order phenotypic signal via unsupervised learning technology, feature learning technology or dimensionality reduction technology.

[0298] 15. The method according to embodiment 1, further comprising:

[0299] (h) access to collections of molecular variants with putative or known phenotypic effects from existing sources;

[0300] (i) using the predictive model to increase the collection of molecular variants with putative or known phenotypic effects;

[0301] (j) selecting a first set of genotypes with putative or known phenotypic effects using a sampling model;

[0302] (k) using a sampling model to select a second set of genotypes with unknown, putative, or known phenotypic effects;

[0303] (l) using a sampling model to select a third set of genotypes with unknown, putative, or known phenotypic effects;

[0304] (m) generating a functional model by applying a statistical learning technique that relates molecular signals, phenotypic signals or population signals of the first set of genotypes to putative or known phenotypic effects;

[0305] (n) applying the functional model based on the molecular signals, phenotypic signals or

[0306] The population signal is predicted to generate predicted phenotypic effects for the second set of genotypes;

[0307] (o) generating an inference model by applying statistical learning techniques, wherein the inference model relates the unmeasured features to the phenotypic effects of the molecular variants; and

[0308] (p) generating predicted phenotypic effects for the third set of genotypes by applying the inference model to make predictions based on unmeasured characteristics of the third set of genotypes.

[0309] 16. The method according to embodiment 1 also includes generating a functional model by applying a statistical learning technique that combines the molecular signal, the phenotypic signal or the population signal with the phenotypic impact of the molecular variant of the functional element, wherein the statistical learning technique is a neural network.

[0310] 17. A method according to embodiment 1, wherein the population signal associated with the molecular variant describes the dynamics of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment in a subpopulation of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment from a unique molecular state associated with the molecular variant.

[0311] 18. The method according to embodiment 1, wherein:

[0312] (i) the molecular score or the phenotypic score of the molecular variant comprises an adjusted molecular score or phenotypic score, which is calculated as the difference between the molecular score or the phenotypic score of the molecular variant and the molecular score or the phenotypic score of a reference molecular variant or a reference single cell, cellular compartment, subcellular compartment or synthetic compartment, or

[0313] (ii) the molecular score or the phenotypic score of the molecular variant comprises an adjusted molecular score or phenotypic score, which is calculated by normalizing the molecular score or the phenotypic score of the molecular variant relative to the molecular score or the phenotypic score of a reference molecular variant or a reference single cell, cellular compartment, subcellular compartment or synthetic compartment.

[0314] 19. The method according to embodiment 1, wherein:

[0315] (i) the molecular signal, phenotypic signal or population signal of the molecular variant comprises an adjusted molecular signal, phenotypic signal or population signal, respectively, which is calculated as the difference between the molecular signal, phenotypic signal or population signal of the molecular variant and the molecular signal, phenotypic signal or population signal of a reference molecular variant;

[0316] (ii) the molecular signal, the phenotypic signal or the population signal associated with the molecular variant comprises an adjusted molecular signal, phenotypic signal or population signal, respectively, which is calculated by normalizing the molecular signal, the phenotypic signal or the population signal associated with the molecular variant with the molecular signal, the phenotypic signal or the population signal of a reference molecular variant; or

[0317] (iii) the molecular signal, the phenotypic signal or the population signal associated with the molecular variant comprises an adjusted molecular signal, phenotypic signal or population signal, respectively, and the adjusted molecular signal, phenotypic signal or population signal is calculated as the quantile of the molecular signal, the phenotypic signal or the population signal associated with the molecular variant among the molecular signal, phenotypic signal or population signal of a reference molecular variant.

[0318] 20. A system comprising:

[0319] (a) storage; and

[0320] (b) at least one processor coupled to the memory and configured to:

[0321] [i] receiving a molecular variant associated with one or more functional elements within a model system, wherein the model system comprises a single cell, a cellular compartment, a subcellular compartment, or a synthetic compartment;

[0322] [ii] determining a molecular score or a phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment,

[0323] wherein the molecular score is a molecular measurement, a molecular process or a molecular signature, or a combination thereof, of a single cell, a cellular compartment, a subcellular compartment or a synthetic compartment, and

[0324] wherein the phenotypic score is derived by applying supervised learning techniques to relate the phenotypic effects of molecular variants within a model system to the molecular score of the model system;

[0325] [iii] determining a molecular signal or a phenotypic signal associated with the molecular variant based on each molecular score or phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment carrying the molecular variant;

[0326] [iv] determining a population signal associated with the molecular variant based on the molecular score or phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment carrying the molecular variant,

[0327] wherein for each individual molecular variant, the population sampling layer produces a measure of the relative representation of cells in each molecular state or a measure of the relative representation of cells in a phenotypic score and is used to provide a population signaling matrix that describes how the molecular variant affects cells at the population level, and

[0328] wherein the population signal matrix contains a plurality of population signals for a plurality of molecular variants;

[0329] [v] determining a functional score or a functional classification of the molecular variant based on statistical learning, wherein the statistical learning correlates the molecular signal, the phenotypic signal or the population signal of the molecular variant with the phenotypic impact of the molecular variant;

[0330] [vi] deriving an evidence score or evidence classification for the molecular variant based on the functional score or functional classification, modeling of the functional score or functional classification, modeling of the predictor score or predictor classification, or modeling of the hotspot score or hotspot classification,

[0331] The training and validation layers use cross-validation techniques for training and quality control, which are then evaluated by the test layer to derive a predictor score for molecular variant classification, and

[0332] wherein a mutation density probability and / or a false discovery rate is defined as a hotspot score and a cutoff is applied to statistically define hotspot classifications to designate residues in candidate clusters; and

[0333] [vii] determining the phenotypic impact of the molecular variant based on the functional score, the functional classification, the evidence score or the evidence classification.

[0334] This manual also includes the following contents:

[0335] 1. A computer-implemented method for determining the phenotypic impact of a molecular variant identified in a biological sample, comprising:

[0336] receiving a molecular variant associated with one or more functional elements within a model system, wherein the model system comprises a single cell, a cellular compartment, a subcellular compartment, or a synthetic compartment;

[0337] determining a molecular score or a phenotypic score of the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment;

[0338] determining a molecular signature or a phenotypic signature associated with a specific molecular variant based on the respective molecular scores or phenotypic scores of the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment carrying the molecular variant;

[0339] determining a population signal associated with a specific molecular variant based on the molecular score or phenotypic score of the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment carrying the molecular variant;

[0340] determining a functional score or a functional classification of the molecular variant based on statistical learning, wherein the statistical learning correlates the molecular signal, the phenotypic signal, or the population signal of the molecular variant with the phenotypic impact of the molecular variant;

[0341] deriving an evidence score or evidence classification for the molecular variant based on the functional score or functional classification, modeling of the functional score or functional classification, modeling of the predictor score or predictor classification, or modeling of the hotspot score or hotspot classification; and

[0342] Based on the functional score, the functional classification, the evidence score, or the evidence classification, the phenotypic impact of the molecular variant is determined.

[0343] 2. A method according to item 1, wherein the evidence score or the evidence classification is determined based on the molecular signal, the phenotypic signal or the population signal of the molecular variant in one or more functional elements.

[0344] 3. The method according to item 1, wherein the evidence score or evidence classification is obtained from the functional score or functional classification, the predictor score or predictor classification, or the hotspot score or hotspot classification.

[0345] 4. A method according to item 1, wherein the evidence score or evidence classification is obtained by applying the statistical learning to associate the evidence score and evidence classification with the phenotypic impact of the molecular variant using regression or classification.

[0346] 5. The method according to item 1, wherein the functional score or functional classification of the molecular variant is obtained by applying statistical learning to associate the molecular signal with the phenotypic effect of the molecular variant using regression or classification.

[0347] 6. The method according to item 4, wherein the phenotypic impact of the molecular variant is derived based on a clinical database, phenotype database, population database, molecular annotation database or functional database of the variant, subject or population.

[0348] 7. The method according to item 4, wherein the phenotypic impact of the molecular variant is derived based on molecular signals such as mutation load, mutation rate and mutation signature.

[0349] 8. A method according to item 1, wherein the functional score or functional classification of the molecular variant is derived from multiple statistical models generated using independent or non-overlapping estimates of the molecular signal, the phenotypic signal or the population signal.

[0350] 9. A method according to Item 1, wherein the functional score or functional classification of the molecular variant is derived from a functional modeling engine (FME), wherein the FME is generated by applying machine learning techniques to associate undetermined features of the molecular variant with the functional score or functional classification, and wherein the undetermined features include evolutionary features, population features, functional features, structural features, kinetic features, and physicochemical features.

[0351] 10. A method according to item 1, wherein the predictor score or predictor classification of the molecular variant is derived from a variant interpretation engine (VIE), wherein the VIE is generated by applying machine learning techniques to associate the functional score or functional classification and the unmeasured features with the phenotypic impact of the molecular variant.

[0352] 11. A method according to item 1, wherein the predictor score or predictor classification is derived from a low-order variant interpretation engine (VIE), wherein the low-order VIE is functional element-specific, functional type-specific or condition-specific.

[0353] 12. A method according to item 1, wherein the predictor score or predictor classification is derived from a high-order variant interpretation engine (VIE), wherein the high-order VIE is pathway-specific, homolog family-specific, enzyme family-specific or condition-specific.

[0354] 13. A method according to item 1, wherein the predictor score or predictor classification is derived from a high-level variant interpretation engine (VIE), wherein the VIE provides information on multiple pathways, homolog families, enzyme families, or conditions.

[0355] 14. A method according to Item 1, wherein the hotspot score or hotspot classification of the molecular variant is derived from significant mutation regions and significant mutation networks (SMR / SMN), and the significant mutation regions and significant mutation networks are calculated by applying spatial clustering technology to detect residue regions and networks with high density of molecular variants with high or low functional scores or specific functional classifications.

[0356] 15. A method according to Item 1, wherein the molecular signal comprises a low-order molecular signal of the molecular variant, and the low-order molecular signal is derived as a summary statistical model, a summary statistical model, a descriptive statistical model, an inferential statistical model or a Bayesian inference model of the molecular score measured in the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment carrying the molecular variant.

[0357] 16. A method according to item 1, wherein the molecular signal comprises a higher-order molecular signal of the molecular variant, and the higher-order molecular signal is derived by applying a pre-existing model that associates low-order molecular signals with activity, alteration, defect or state of regulation, signal transduction, pathway, processing or cell cycle.

[0358] 17. The method according to item 1, wherein the molecular signal comprises a high-order molecular signal of the molecular variant, and the high-order molecular signal is derived from a low-order molecular signal via unsupervised learning technology, feature learning technology or dimensionality reduction technology.

[0359] 18. A method according to Item 1, wherein the molecular signal comprises a low-order molecular score corresponding to a molecular measurement result, a molecular process, or a molecular feature from the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment.

[0360] 19. A method according to item 1, wherein the molecular signal comprises a high-order molecular score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment, and the high-order molecular score is obtained by applying a pre-existing model that associates low-order molecular scores with activity, alteration, defect or state of regulation, signal transduction, pathway, processing or cell cycle.

[0361] 20. A method according to Item 1, wherein the molecular signal comprises a high-order molecular score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment, and the high-order molecular score is derived from the low-order molecular score via unsupervised learning technology, feature learning technology or dimensionality reduction technology.

[0362] 21. A method according to item 20, wherein an autoencoder neural network is trained to learn a compressed representation of low-order molecular scores, and the autoencoder is used to encode low-order molecular signals into high-order compressed representations.

[0363] 22. A method according to item 21, wherein the autoencoder is trained as a denoising autoencoder (DAE), or the autoencoder is constructed as a neural network with a fully connected layer, or the autoencoder is constructed as a neural network with a symmetric number of neurons, or the autoencoder is built with a rectified linear unit (ReLu) for activation, or the autoencoder is trained using an Adam optimizer, or the autoencoder is cell type-specific, gene-specific, pathway-specific or disorder-specific.

[0364] 23. A method according to item 18, wherein the molecular measurement results correspond to site-specific measurements of gene expression, protein expression, chromatin accessibility, epigenetic modification, regulatory activity, post-transcriptional processing, post-translational modification, mutation status, mutation load or mutation rate of molecules within the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment.

[0365] 24. A method according to item 18, wherein the molecular process corresponds to multi-site measurements of gene expression, protein expression, chromatin accessibility, epigenetic modification, regulatory activity, transcriptional activity, translational activity, signaling activity, pathway activity, mutation status, mutation load or mutation rate, etc. derived from molecular measurements within the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment.

[0366] 25. A method according to item 18, wherein the molecular features correspond to global measurements of gene expression, protein expression, chromatin accessibility, epigenetic modification, regulatory activity, transcriptional activity, translational activity, signaling activity, pathway activity, mutation status, mutation load or mutation rate, etc., derived from molecular measurements or molecular processes within the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment.

[0367] 26. The method according to item 18, wherein the molecular measurement results are obtained by applying single-cell barcoding technology and nucleic acid sequencing technology to a population of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment.

[0368] 27. A method according to item 18, wherein the molecular measurement results may include: sequencing read quality control, cellular barcode identification or quality control, molecular barcode identification or quality control, sequencing read alignment with a reference genome, sequencing read alignment filtering or quality control, mapping filtered and quality controlled sequencing reads to functional elements, mapping filtered and quality controlled molecular barcodes to functional elements, and mapping filtered and quality controlled sequencing reads or molecular barcodes of specific cellular barcodes to functional elements.

[0369] 28. A method according to Item 1, wherein the molecular signal, the phenotypic signal or the population signal is molecular state specific, which originates from a population of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment in a specific molecular state to allow learning in a state-specific learning layer.

[0370] 29. A method according to Item 1, wherein the molecular signal, the phenotypic signal or the population signal is molecular state agnostic, which is derived from a population of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment in multiple molecular states to allow learning in a state-agnostic learning layer.

[0371] 30. A method according to Item 1, wherein the molecular signal, the phenotypic signal or the population signal is molecular state ordered, which originates from a population of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment in multiple molecular states to allow learning in a multi-state learning layer.

[0372] 31. A method according to item 1, wherein the molecular state of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment is derived by applying a pre-existing model that associates a molecular score or a phenotypic score with the molecular state, wherein the model assigns single cells to cell cycle stages based on previously characterized gene expression signatures.

[0373] 32. A method according to Item 1, wherein the molecular state of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment is obtained by applying unsupervised learning technology, feature learning technology or dimensionality reduction technology to the molecular score or phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment.

[0374] 33. A method according to Item 1, wherein the molecular signal, the phenotypic signal or the population signal is calculated from independent or non-overlapping populations of single cells, cellular compartments, subcellular compartments or synthetic compartments, and the independent or non-overlapping populations are selected by random sampling from the single cells, the cellular compartments, the subcellular compartments or the synthetic compartments carrying the same molecular variant.

[0375] 34. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant within a functional element, gene or pathway associated with a Mendelian disease.

[0376] 35. A method according to item 1, wherein the molecular variants correspond to coding or non-coding variants within functional elements, genes and pathways associated with known cancer driver genes.

[0377] 36. A method according to item 1, wherein the molecular variants correspond to coding or non-coding variants, which are within functional elements, genes and pathways associated with drug response variation.

[0378] 37. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within a previously identified mutation hotspot of functional elements, genes and pathways associated with other clinically valuable genes.

[0379] 38. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant within a previously identified mutation hotspot of functional elements, genes and pathways associated with Mendelian disease.

[0380] 39. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant within previously identified mutation hotspots of functional elements, genes and pathways associated with known cancer driver genes.

[0381] 40. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within previously identified mutation hotspots of functional elements, genes and pathways associated with drug response variation.

[0382] 41. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within a previously identified mutation hotspot of functional elements, genes and pathways associated with other clinically valuable genes.

[0383] 42. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 10bp of previously identified mutation hotspots of functional elements, genes and pathways associated with Mendelian diseases.

[0384] 43. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant that is within 10bp of previously identified mutation hotspots of functional elements, genes and pathways associated with known cancer driver genes.

[0385] 44. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 10bp of previously identified mutation hotspots of functional elements, genes and pathways associated with drug response variation.

[0386] 45. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 10bp of previously identified mutation hotspots of functional elements, genes and pathways associated with other genes of clinical value.

[0387] 46. ​​A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant that is within 50 bp of previously identified mutation hotspots of functional elements, genes and pathways associated with Mendelian diseases.

[0388] 47. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant that is within 50bp of previously identified mutation hotspots of functional elements, genes and pathways associated with known cancer driver genes.

[0389] 48. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 50bp of previously identified mutation hotspots of functional elements, genes and pathways associated with drug response variation.

[0390] 49. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant that is within 50 bp of previously identified mutation hotspots of functional elements, genes and pathways associated with other genes of clinical value.

[0391] 50. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant that is within 100 bp of previously identified mutation hotspots of functional elements, genes and pathways associated with Mendelian diseases.

[0392] 51. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant that is within 100bp of previously identified mutation hotspots of functional elements, genes and pathways associated with known cancer driver genes.

[0393] 52. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 100bp of previously identified mutation hotspots of functional elements, genes and pathways associated with drug response variation.

[0394] 53. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 100bp of previously identified mutation hotspots of functional elements, genes and pathways associated with other genes of clinical value.

[0395] 54. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 500bp of previously identified mutation hotspots of functional elements, genes and pathways associated with Mendelian diseases.

[0396] 55. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 500bp of previously identified mutation hotspots of functional elements, genes and pathways associated with known cancer driver genes.

[0397] 56. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 500bp of previously identified mutation hotspots of functional elements, genes and pathways associated with drug response variation.

[0398] 57. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 500bp of previously identified mutation hotspots of functional elements, genes and pathways associated with other genes of clinical value.

[0399] 58. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant that is within 1,000 bp of previously identified mutation hotspots of functional elements, genes and pathways associated with Mendelian diseases.

[0400] 59. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant that is within 1,000 bp of previously identified mutation hotspots of functional elements, genes and pathways associated with known cancer driver genes.

[0401] 60. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant that is within 1,000 bp of previously identified mutation hotspots of functional elements, genes, and pathways associated with drug response variation.

[0402] 61. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant that is within 1,000 bp of previously identified mutation hotspots of functional elements, genes and pathways associated with other clinically valuable genes.

[0403] 62. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant within a previously identified restricted region of functional elements, genes and pathways associated with Mendelian disease.

[0404] 63. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant within a previously identified restricted region of functional elements, genes and pathways associated with known cancer driver genes.

[0405] 64. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within a previously identified restricted region of functional elements, genes and pathways associated with drug response variation.

[0406] 65. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant within a previously identified restricted region of functional elements, genes and pathways associated with other genes of clinical value.

[0407] 66. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 10bp of a previously identified restricted region of functional elements, genes and pathways associated with Mendelian disease.

[0408] 67. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 10bp of a previously identified restricted region of functional elements, genes and pathways associated with known cancer driver genes.

[0409] 68. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 10bp of a previously identified restricted region of functional elements, genes and pathways associated with drug response variation.

[0410] 69. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 10bp of a previously identified restricted region of functional elements, genes and pathways associated with other genes of clinical value.

[0411] 70. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 50bp of a previously identified restricted region of functional elements, genes and pathways associated with Mendelian disease.

[0412] 71. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 50bp of a previously identified restricted region of functional elements, genes and pathways associated with known cancer driver genes.

[0413] 72. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 50bp of a previously identified restricted region of functional elements, genes and pathways associated with drug response variation.

[0414] 73. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 50bp of a previously identified restricted region of functional elements, genes and pathways associated with other genes of clinical value.

[0415] 74. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 100bp of a previously identified restricted region of functional elements, genes and pathways associated with Mendelian disease.

[0416] 75. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 100bp of a previously identified restricted region of functional elements, genes and pathways associated with known cancer driver genes.

[0417] 76. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 100bp of a previously identified restricted region of functional elements, genes and pathways associated with drug response variation.

[0418] 77. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 100bp of a previously identified restricted region of functional elements, genes and pathways associated with other genes of clinical value.

[0419] 78. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 500bp of a previously identified restricted region of functional elements, genes and pathways associated with Mendelian disease.

[0420] 79. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 500bp of a previously identified restricted region of functional elements, genes and pathways associated with known cancer driver genes.

[0421] 80. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 500bp of a previously identified restricted region of functional elements, genes and pathways associated with drug response variation.

[0422] 81. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 500bp of a previously identified restricted region of functional elements, genes and pathways associated with other genes of clinical value.

[0423] 82. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 1,000bp of a previously identified restricted region of functional elements, genes and pathways associated with Mendelian disease.

[0424] 83. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 1,000bp of a previously identified restricted region of functional elements, genes and pathways associated with known cancer driver genes.

[0425] 84. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 1,000bp of a previously identified restricted region of functional elements, genes and pathways associated with drug response variation.

[0426] 85. A method according to item 1, wherein the molecular variant corresponds to a coding or non-coding variant, which is within 1,000bp of a previously identified restricted region of functional elements, genes and pathways associated with other genes of clinical value.

[0427] 86. A method according to item 1, wherein the phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment represents the phenotypic association of the molecular variant identified within the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment.

[0428] 87. A method according to item 1, wherein the phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment comprises a low-order phenotypic score, wherein the low-order phenotypic score corresponds to a score or classification generated by a phenotypic model by using a statistical learning technique, which associates the molecular scores and molecular states of the model system with the phenotypic effects of molecular variants within each model system.

[0429] 88. A method according to item 87, wherein the phenotypic model is generated using a neural network architecture for single-task or multi-task statistical learning, which associates molecular scores from one or more functional elements with one or more phenotypic effects of molecular variants in the one or more functional elements.

[0430] 89. A method according to item 1, wherein the phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment comprises a high-order phenotypic score, wherein the high-order phenotypic score is derived by applying a pre-existing model that associates low-order phenotypic scores with activity, alteration, defect or state of regulation, signal transduction, pathway, processing or cell cycle.

[0431] 90. A method according to Item 1, wherein the phenotypic score of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment comprises a high-order phenotypic score, wherein the high-order phenotypic score is derived from a low-order phenotypic score via unsupervised learning techniques, feature learning techniques or dimensionality reduction techniques.

[0432] 91. A method according to Item 1, wherein the phenotypic signal associated with the molecular variant comprises a low-order phenotypic signal associated with the molecular variant, wherein the low-order phenotypic signal associated with the molecular variant is derived as a summary statistical model, a descriptive statistical model, an inferential statistical model or a Bayesian inference model of the phenotypic score measured in the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment carrying the molecular variant.

[0433] 92. A method according to Item 1, wherein the phenotypic signal associated with the molecular variant comprises a high-order phenotypic signal associated with the molecular variant, wherein the high-order phenotypic signal associated with the molecular variant is derived by applying a pre-existing model that associates low-order phenotypic signals with activity, alteration, defect or state of regulation, signal transduction, pathway, processing or cell cycle.

[0434] 93. A method according to Item 1, wherein the phenotypic signal associated with the molecular variant comprises a high-order phenotypic signal associated with the molecular variant, wherein the high-order phenotypic signal associated with the molecular variant is derived from a low-order phenotypic signal via unsupervised learning techniques, feature learning techniques or dimensionality reduction techniques.

[0435] 94. The method according to item 1, further comprising:

[0436] Access to collections of molecular variants with putative or known phenotypic effects from existing sources;

[0437] using a predictive model to increase the collection of molecular variants with putative or known phenotypic effects;

[0438] A first set of genotypes with putative or known phenotypic effects is selected using a sampling model;

[0439] A second set of genotypes with unknown, putative, or known phenotypic effects was selected using a sampling model;

[0440] A third set of genotypes with unknown, putative, or known phenotypic effects was selected using a sampling model;

[0441] generating a functional model by applying a statistical learning technique that relates molecular signals, phenotypic signals, or population signals of the first set of genotypes to putative or known phenotypic effects;

[0442] generating predicted phenotypic effects for the second set of genotypes by applying the functional model to make predictions based on molecular signals, phenotypic signals, or population signals for the second set of genotypes;

[0443] generating an inference model by applying statistical learning techniques, wherein the inference model relates the unmeasured features to the phenotypic effects of the molecular variants; and

[0444] Predicted phenotypic effects are generated for the third set of genotypes by applying the inference model to make predictions based on unmeasured characteristics of the third set of genotypes.

[0445] 95. A method according to item 94, wherein the predictive model is a gene-specific, domain-specific, homolog-specific or genome-wide computational predictor or functional assay.

[0446] 96. A method according to item 94, wherein the predictive model provides a performance or confidence estimate for each prediction of the predictive model.

[0447] 97. A method according to item 94, wherein the positive predictive value (PPV) of the predictive model comprises a function of the performance or confidence estimate of the prediction of the predictive model.

[0448] 98. A method according to item 94, wherein the negative predictive value (NPV) of the predictive model comprises a function of the performance or confidence estimate of the prediction of the predictive model.

[0449] 99. A method according to item 94, wherein the predictive model is a molecular impact predictor.

[0450] 100. The method according to item 94, wherein the prediction model predicts that a premature termination molecular variant, a nonsense molecular variant or a truncated molecular variant in a protein encoding functional element is a loss-of-function variant.

[0451] 101. A method according to item 94, wherein the prediction model predicts that synonymous or silent molecular variants in protein-coding functional elements are neutral variants.

[0452] 102. The method according to item 1, further comprising:

[0453] A functional model is generated by applying a statistical learning technique that combines the molecular signal, the phenotypic signal or the population signal with the phenotypic effects of the molecular variants of the functional element.

[0454] 103. The method according to item 102, wherein generating the functional model further comprises:

[0455] The functional model is generated using a neural network architecture for single-task or multi-task learning, which associates the molecular signal, the phenotypic signal or the population signal from the functional element with the one or more phenotypic effects of the molecular variants of the functional element.

[0456] 104. The method according to item 1, further comprising:

[0457] A phenotypic model is generated by applying a statistical learning technique that combines the molecular scores with the phenotypic effects of the molecular variants of the functional elements.

[0458] 105. The method according to item 104, wherein generating a phenotypic model further comprises:

[0459] A phenotypic model is generated using a neural network architecture for single-task or multi-task learning that associates the molecular scores from the functional elements with the one or more phenotypic effects of the molecular variants of the functional elements.

[0460] 106. The method according to item 1, further comprising:

[0461] introducing the molecular variant into the functional element within the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment;

[0462] identifying said molecular variant within said single cell, said cellular compartment, said subcellular compartment, or said synthetic compartment;

[0463] determining the phenotypic impact of the molecular variant within the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment; and

[0464] Determine a molecular measurement, a molecular signature, or a molecular process within the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment.

[0465] 107. A method according to Item 1, wherein the population signal associated with the molecular variant describes the distribution of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment associated with the molecular variant in a subpopulation of single cells, cellular compartments, subcellular compartments or synthetic compartments from a unique molecular state.

[0466] 108. A method according to Item 1, wherein the population signal associated with the molecular variant describes the dynamics of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment associated with the molecular variant in a subpopulation of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment from a unique molecular state.

[0467] 109. A method according to Item 1, wherein the population signal associated with the molecular variant describes the distribution change of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment in a subpopulation of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment from a unique molecular state associated with the molecular variant.

[0468] 110. A method according to Item 1, wherein the population signal associated with the molecular variant describes the dynamics of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment in a subpopulation of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment from a unique molecular state associated with the molecular variant.

[0469] 111. A method according to item 107, wherein based on the molecular score or the phenotypic score, a clustering technique is applied to cluster and assign the single cells, the cellular compartments, the subcellular compartments or the synthetic compartments.

[0470] 112. A method according to item 111, wherein a Gaussian mixture model (GMM) is applied to cluster the single cells, the cellular compartments, the subcellular compartments or the synthetic compartments and assign them to a limited number of molecular states.

[0471] 113. A method according to item 111, wherein a variational Gaussian mixture model (VGMM) is applied using a Dirichlet process to cluster the single cells, the cellular compartments, the subcellular compartments or the synthetic compartments and assign them to a presumed number of molecular states.

[0472] 114. A method according to item 107, wherein the population signal associated with the molecular variant is determined as the ratio of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment associated with the molecular variant corresponding to a specific molecular state.

[0473] 115. A method according to Item 1, wherein the molecular score or the phenotypic score of the molecular variant comprises an adjusted molecular score or phenotypic score, and the adjusted molecular score or phenotypic score is calculated as the difference between the molecular score or the phenotypic score of the molecular variant and the molecular score or the phenotypic score of a reference molecular variant or a reference single cell, cellular compartment, subcellular compartment or synthetic compartment.

[0474] 116. A method according to Item 1, wherein the molecular score or the phenotypic score of the molecular variant comprises an adjusted molecular score or phenotypic score, and the adjusted molecular score or phenotypic score is calculated by normalizing the molecular score or the phenotypic score of the molecular variant relative to the molecular score or phenotypic score of a reference molecular variant or a reference single cell, cellular compartment, subcellular compartment or synthetic compartment.

[0475] 117. The method according to Item 1, wherein the molecular signal, phenotypic signal or population signal of the molecular variant comprises an adjusted molecular signal, phenotypic signal or population signal, respectively, and the adjusted molecular signal, phenotypic signal or population signal is calculated as the difference between the molecular signal, phenotypic signal or population signal of the molecular variant and the molecular signal, phenotypic signal or population signal of the reference molecular variant.

[0476] 118. The method according to Item 1, wherein the molecular signal, the phenotypic signal or the population signal associated with the molecular variant comprises an adjusted molecular signal, phenotypic signal or population signal, respectively, and the adjusted molecular signal, phenotypic signal or population signal is calculated by normalizing the molecular signal, the phenotypic signal or the population signal associated with the molecular variant with the molecular signal, phenotypic signal or population signal of a reference molecular variant.

[0477] 119. The method according to Item 1, wherein the molecular signal, the phenotypic signal or the population signal associated with the molecular variant respectively comprises an adjusted molecular signal, phenotypic signal or population signal, and the adjusted molecular signal, phenotypic signal or population signal is calculated as the quantile of the molecular signal, the phenotypic signal or the population signal associated with the molecular variant among the molecular signal, phenotypic signal or population signal of a reference molecular variant.

[0478] 120. A computer-implemented method further comprising:

[0479] The first set of genotypes with phenotypic effects is selected;

[0480] A second set of genotypes with phenotypic effects were selected;

[0481] obtaining molecules from a first number of single cells, cellular compartments, subcellular compartments, or synthetic compartments associated with the first set of genotypes using single cell capture or barcoding technology;

[0482] obtaining a first number of reads of molecular reads for each model system by performing sequencing, sequencing read quality control, cellular barcode identification or quality control, molecular barcode identification or quality control, sequencing read alignment to a reference genome, or read alignment filtering or quality control using the model system associated with the first set of genotypes;

[0483] obtaining molecules from a second number of cells of the single cells, the cellular compartments, the subcellular compartments, or the synthetic compartments associated with the first set of genotypes using single cell capture or barcoding technology;

[0484] obtaining a second number of reads of molecular reads per model by performing sequencing, sequencing read quality control, cellular barcode identification or quality control, molecular barcode identification or quality control, sequencing read alignment to a reference genome, or read alignment filtering or quality control using the model system associated with the first set of genotypes;

[0485] Total molecular reads or total molecular measurements of molecular reads from the total number of reads per model system derived from single cells, cellular compartments, subcellular compartments, or synthetic compartments for the total number of cells per genotype;

[0486] Determining a molecular score, a phenotypic score, a molecular signal, a phenotypic signal, or a population signal for the first set of genotypes using the total molecular reads and the total molecular measurements by applying a statistical learning technique for feature selection or dimensionality reduction to generate an overall reduced dimensionality model;

[0487] generating an overall functional model using the total molecular reads and the total molecular measurements by applying a statistical learning technique that relates molecular signals, phenotypic signals, or population signals from the total dimensionality reduction model to the phenotypic effects of the first set of genotypes;

[0488] determining a threshold performance of a functional score or functional classification using the total number of cells, the total number of reads, the total dimensionality reduction model, or the total functional model for predicting the phenotypic impact of the first set of genotypes;

[0489] deriving the best molecular reads or the best molecular measurements of the molecular reads from the best number of reads for each model system from the single cells, cellular compartments, subcellular compartments or synthetic compartments of the best number of cells for each genotype, wherein the best molecular reads and the best molecular measurements are obtained by subsampling the total molecular reads or the total molecular measurements;

[0490] generating an optimal dimensionality reduction model by applying a statistical learning technique for feature selection or dimensionality reduction to determine a molecular score, a phenotypic score, a molecular signal, a phenotypic signal, or a population signal for the first set of genotypes using the optimal molecular reads and the optimal molecular measurements;

[0491] generating an optimal functional model using the optimal molecular reads and the optimal molecular measurements by applying a statistical learning technique that relates a molecular signal, a phenotypic signal, or a population signal from the optimal dimensionality reduction model to the phenotypic effects of the first set of genotypes;

[0492] validating a threshold performance of the functional score or functional classification based on the optimal number of cells, the optimal number of reads, the optimal dimensionality reduction model, or the optimal functional model for predicting the phenotypic impact of the first set of genotypes;

[0493] obtaining molecules from said optimal number of single cells, cellular compartments, subcellular compartments, or synthetic compartments associated with said second set of genotypes using single cell capture or barcoding technology;

[0494] Obtaining the optimal number of molecular reads for each model system by performing sequencing, sequencing read quality control, cellular barcode identification or quality control, molecular barcode identification or quality control, sequencing read alignment to a reference genome, or read alignment filtering or quality control using the model system associated with the second set of genotypes; and

[0495] A functional score or a functional classification is generated for the second set of genotypes based on the optimal number of cells, the optimal number of reads, the optimal dimensionality reduction model, or the optimal functional model.

[0496] 121. A computer-implemented method for scoring the phenotypic impact of a molecular variant, comprising:

[0497] evaluating the evidence dataset based on the accuracy of the evidence dataset;

[0498] verifying the evidence dataset based on the accuracy of the evidence dataset;

[0499] Optimizing the evidence dataset based on the accuracy of the evidence dataset; and

[0500] The phenotypic impact of the molecular variant is determined based on the evaluation, validation and optimization of the evidence set.

[0501] 122. A method according to item 121, wherein the evidence dataset comprises functional scores or functional classifications of molecular variants based on a machine learning model, wherein the machine learning model associates the molecular signal, phenotypic signal or population signal of the molecular variant with the phenotypic impact of the molecular variant.

[0502] 123. A method according to item 121, wherein the evidence dataset comprises predictor scores or predictor classifications from genome-wide, homolog-specific, enzyme class-specific, domain-specific or gene-specific computational predictors.

[0503] 124. A method according to item 121, wherein the evidence dataset comprises hotspot scores or hotspot classifications from mutation hotspots.

[0504] 125. A method according to item 121, wherein the evidence dataset comprises population scores or population classifications derived from variant classifications based on population genomics metrics.

[0505] 126. The method according to item 121, further comprising:

[0506] An evaluation metric is calculated to assess the consistency between the evidence dataset and the functional score or functional classification.

[0507] 127. The method according to item 121, wherein the evaluation indicators include Pearson correlation coefficient, Spearman rank correlation, Kendall correlation, Matthew correlation coefficient, Cohen's kappa coefficient, Youden index, F measure, true positive rate, true negative rate, positive predictive value, negative predictive value, positive likelihood ratio, negative likelihood ratio or diagnostic odds ratio.

[0508] 128. A method according to item 121, wherein the verification of the evidence dataset includes verifying the evidence dataset based on the evaluation metric.

[0509] 129. A method according to item 121, wherein the optimization of the evidence dataset includes selecting or removing data within the evidence dataset based on the evaluation metric.

[0510] 130. A computer-implemented method for scoring the phenotypic impact of a molecular variant, comprising:

[0511] evaluating the evidence dataset based on inherent biases of the evidence dataset;

[0512] validating the evidence dataset based on the intrinsic bias of the evidence dataset;

[0513] optimizing the evidence dataset based on the intrinsic bias of the evidence dataset; and

[0514] A score for the phenotypic impact of the molecular variant is determined based on the evaluation, validation and optimization of the evidence set.

[0515] 131. A method according to item 130, wherein the deviation of the evidence dataset is measured as the statistical distance between the observed evidence score or evidence classification of the variant in the evidence dataset and the expected evidence score or evidence classification of the variant in the reference dataset.

[0516] 132. A method according to item 130, wherein the confirmation bias of the evidence dataset is measured as the statistical distance between the observed characteristics and properties of the variants in the evidence dataset and the expected characteristics and properties of the variants in the reference dataset defined based on matching quantiles or classifications.

[0517] 133. A method according to item 130, wherein the confirmation bias of the evidence dataset is measured as the statistical distance between the observed features and properties of the variants in the evidence dataset and the expected features and properties of the variants in the reference dataset defined based on a matching distribution of evidence scores or evidence classifications.

[0518] 134. A method according to item 130, wherein the verification of the evidence dataset includes verifying the evidence dataset based on a target evaluation bias indicator.

[0519] 135. A method according to item 130, wherein the optimization of the evidence dataset includes selecting or removing data within the evidence dataset based on target validation criteria.

[0520] 136. A system comprising:

[0521] Memory; and

[0522] at least one processor coupled to the memory and configured to:

[0523] receiving a molecular variant associated with one or more functional elements within a model system, wherein the model system comprises a single cell, a cellular compartment, a subcellular compartment, or a synthetic compartment;

[0524] determining a molecular score or a phenotypic score of the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment;

[0525] determining a molecular signature or a phenotypic signature associated with a specific molecular variant based on each molecular signature or a phenotypic score of the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment;

[0526] determining a population signal associated with a specific molecular variant based on the molecular score or phenotypic score of the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment;

[0527] determining a functional score or a functional classification of the molecular variant based on statistical learning, wherein the statistical learning correlates the molecular signal, the phenotypic signal or the population signal of the molecular variant with the phenotypic impact of the molecular variant;

[0528] deriving an evidence score or evidence classification for the molecular variant based on the functional score or functional classification, modeling of the functional score or functional classification, modeling of the predictor score or predictor classification, or modeling of the hotspot score or hotspot classification;

[0529] Based on the functional score, the functional classification, the evidence score, or the evidence classification, the phenotypic impact of the molecular variant is determined.

[0530] 137. A tangible computer readable device having instructions stored thereon, which, when executed by at least one computing device, cause the at least one computing device to perform operations comprising:

[0531] receiving a molecular variant associated with one or more functional elements within a model system, wherein the model system comprises a single cell, a cellular compartment, a subcellular compartment, or a synthetic compartment;

[0532] determining a molecular score or a phenotypic score of the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment;

[0533] determining a molecular signature or a phenotypic signature associated with a specific molecular variant based on the respective molecular scores or phenotypic scores of the single cell, the cellular compartment, the subcellular compartment or the synthetic compartment carrying the molecular variant;

[0534] determining a population signal associated with a specific molecular variant based on the molecular score or phenotypic score of the single cell, the cellular compartment, the subcellular compartment, or the synthetic compartment;

[0535] determining a functional score or a functional classification of the molecular variant based on statistical learning, wherein the statistical learning correlates the molecular signal, the phenotypic signal or the population signal of the molecular variant with the phenotypic impact of the molecular variant;

[0536] deriving an evidence score or evidence classification for the molecular variant based on the functional score or functional classification, modeling of the functional score or functional classification, modeling of the predictor score or predictor classification, or modeling of the hotspot score or hotspot classification;

[0537] Based on the functional score, the functional classification, the evidence score, or the evidence classification, the phenotypic impact of the molecular variant is determined.

Claims

1. A method for determining the phenotypic impact of a target molecular variant, the method comprising: Receive multiple samples, wherein the plurality of samples comprises a plurality of molecular variants, and each sample comprises a variant in a gene, The multiple molecular variants are divided into two groups: (a) contains the true set of molecular variants with known phenotypic effects, and (b) a target set comprising molecular variants with unknown phenotypic effects, wherein the target set comprises target molecular variants; Training machine learning models using known associations between molecular variants in the true set and known phenotypic effects, wherein the known association is based on a plurality of dependency features determined using a functional assay that produces or is derived from a molecular measurement for each molecular variant in the true set; and Use trained machine learning models to determine the phenotypic impact of molecular variants of interest.

2. The method of claim 1, wherein the plurality of samples comprises single cells, cellular compartments, subcellular compartments, or synthetic compartments.

3. The method of claim 1, wherein the plurality of molecular variants comprises coding or non-coding variants within previously identified mutation hotspots of functional elements, genes and pathways associated with other genes of clinical value, mutation hotspots of functional elements, genes and pathways associated with Mendelian diseases, pathways associated with known cancer driver genes, or pathways associated with drug response variants.

4. The method of claim 1, wherein the plurality of molecular variants are derived based on a clinical database, a phenotype database, a population database, a molecular annotation database, or a functional database of variants, subjects, or populations or are generated using a mutagenesis assay.

5. The method of claim 1, wherein the known phenotypic effects of the molecular variants in the true set and the unknown phenotypic effects of the target molecular variants in the target set measure the pathogenicity, function or relative effect of the molecular variants.

6. The method of claim 1, wherein the molecular measurements further comprise site-specific measurements of gene expression, protein expression, chromatin accessibility, epigenetic modification, regulatory activity, post-transcriptional processing, post-translational modification, mutational status, mutational load, or mutation rate of a molecule within each sample in the plurality of samples.

7. The method of claim 1, wherein the machine learning model is a supervised learning model.

8. The method of claim 1, wherein the results derived from the molecular measurements are generated using a plurality of artificial neural networks (ANNs), wherein the plurality of ANNs comprises: (a) a first ANN that generates a database of molecular measurements for the truth set, (b) a second ANN that uses statistical methods to generate a plurality of associations between each molecular measurement in the database and one or more molecular measurements in the group consisting of molecular states, phenotypes, and genomic indicators, and (c) a third ANN that generates results derived from the molecular measurements by reducing dimensionality and removing noise from associations corresponding to the molecular measurements, wherein the results derived from the molecular measurements are used to determine the phenotypic impact of the target molecular variant.

9. The method according to claim 1, wherein the known association is based on multiple independent features that are not determined for every molecular variant in the true set, and Wherein the plurality of independent features comprises one or more of the following: evolutionary, population, annotation-based, structural, kinetic, physicochemical characteristics, genomic coordinates, transcript coordinates, translation coordinates, and amino acids associated with the variant.

10. The method of claim 1, wherein the method is used to inform a subject of the lifetime risk of developing cancer, wherein the subject has a molecular variant of interest.

11. The method of claim 1, wherein the method is used to identify significantly mutated regions and significantly mutated networks by identifying phenotype-associated mutation densities.

12. A system for determining the phenotypic impact of a target molecular variant, the system comprising: at least one computer hardware processor; and at least one non-transitory computer-readable storage medium storing processor-executable instructions that, when executed by the at least one computer hardware processor, cause the at least one computer hardware processor to perform: Training machine learning models using known associations between molecular variants in the true set and known phenotypic effects, wherein the known association is based on a plurality of dependent features determined using a functional assay that produces or is derived from a molecular measurement for each sample in the true set; and Use trained machine learning models to determine the phenotypic impact of molecular variants of interest.

13. The system of claim 12, wherein the known phenotypic effects of the molecular variants in the true set and the phenotypic effects of the molecular variants of interest measure the pathogenicity, function, or relative effect of the molecular variants.

14. The system of claim 12, wherein the molecular measurements further comprise site-specific measurements of gene expression, protein expression, chromatin accessibility, epigenetic modification, regulatory activity, post-transcriptional processing, post-translational modification, mutational status, mutational load, or mutation rate of a molecule within each sample in the plurality of samples.

15. The system of claim 12, wherein the machine learning model is a supervised learning model.

16. The system of claim 12, wherein the results derived from the molecular measurements are generated using a plurality of artificial neural networks (ANNs), wherein the plurality of ANNs comprises: (a) a first ANN that generates a database of molecular measurements for the truth set, (b) a second ANN that uses statistical methods to generate a plurality of associations between each molecular measurement in the database and one or more molecular measurements in the group consisting of molecular states, phenotypes, and genomic indicators, and (c) a third ANN that generates results derived from the molecular measurements by reducing dimensionality and removing noise from associations corresponding to the molecular measurements, wherein the results derived from the molecular measurements are used to determine the phenotypic impact of the target molecular variant.

17. The system according to claim 12, wherein the known association is based on multiple independent features that are not measured for every sample in the true set, and The plurality of independent features comprises one or more of the following: evolutionary, population, annotation-based, structural, kinetic, physicochemical characteristics, genomic coordinates, transcript coordinates, translation coordinates, and amino acids associated with the variant.

18. The system of claim 12, wherein the system is used to inform a subject of the lifetime risk of developing cancer, wherein the subject has a molecular variant of interest.

19. The system of claim 12, wherein the system is used to identify significantly mutated regions and significantly mutated networks by identifying phenotype-associated mutation density.

20. At least one non-transitory computer-readable storage medium storing processor-executable instructions that, when executed by at least one computer hardware processor, cause the at least one computer hardware processor to perform: Training machine learning models using known associations between molecular variants in the true set and known phenotypic effects, wherein the known association is based on a plurality of dependent features determined using a functional assay that produces or is derived from a molecular measurement for each sample in the true set; and Use trained machine learning models to determine the phenotypic impact of molecular variants of interest.

Citation Information

Patent Citations

  • Systems and Methods for Multi-Scale, Annotation-Independent Detection of Functionally-Diverse Units of Recurrent Genomic Alteration

    US20160378915A1